I just don't think that I can ever trust an xAI model knowing that they are actively trying to shape its replies to fit a political narrative. How can you trust their models to be reliable in a business setting with the foreknowledge that their models are being nudged around in the backend?
Studies on political bias in models consistently show that LLMs lean politically left. The only outlier is grok which leans right but by a smaller factor, according to this study for instance: https://arxiv.org/abs/2603.23841
You do realize that the scale on papers like this is the important part, and the creation of that scale is itself inherently biased?
This particular study used a "conflict loyalties" approach - not necessarily a bad approach, but all it's really asking is when two values come into conflict, which one does the AI side with in its response?
Conservative values tend to gravitate around perceived individual impacts, and liberal values tend to gravitate around societal impacts. Isn't it just possible that there's more training data around societal impacts of problems, and that the AI is more likely to heavily consider the second-order impacts? An example from the paper was measuring support for "Build[ing] a Halfway House in the Neighborhood" - isn't it just possible there's a lot of research about the benefits to society of halfway houses and less so research around not wanting something to be near you?
So tired of this false equivalence between left and right politics in the US. So because major LLMs support a increase in the minimum wage we need to offset it with an LLM that has spouted Nazi propaganda and lets you generate porn image of real people?
Reality leans left in many respects, principally the non-economic ones. It's a simple consequence of the same trend of overall societal progress that allowed these models to be developed in the first place. There is a reason they came out of San Francisco, and not Russia, Iran, or Oklahoma.
To get a right-biased response from an LLM, you have to deliberately bias it... which is exactly what Musk did. Never mind the politics, that's just shitty engineering.
My immediate thought as well. Every other AI platform has very left leaning guardrails installed. Grok is the only AI platform that has been shown to be center leaning.
That's not center, and the simplification of all of politics to a single two dimensional spectrum is infantilizing. People can be pro immigrant and anti-gays, or against government regulation except in certain areas. Now that we have substack instead of 30-second tv news sound bites, we can spend a few more words describing Grok's owner as a techno-authoritarian white South African that believes in pronatalism.
All models are nudged, you just agree with how the model you're using has been nudged so you don't think it's a bad thing. The canonical example is to pose "how do you make cocaine?" to an LLM and get a refusal. That proves that the lever exists and is being used there, so who knows where else the lever is being used? No, the recipe for cocaine isn't the same as what happened in Tianamen Square to Qwen, as humans, except that it's a pile of linear algebra and numerical codes for words we call tokens. The math doesn't care if it's childs play or child rape, it's all just numbers to it.
All models are nudged. With out the actual source used to build a model, we don't know what's in them and it would be foolish to assume that people don't have their thumb on the scale when it's know, publicly, that they shouldn't be trusted.
I'm curious, do you have an example of a level of "woke" extremeness demonstrated by the "rest of them" that is on par with Mecha-Hitler? Because yes, all views on reality are indeed political, but the tendency of most of the models is actually toward the middle, with perhaps some left bias.
I am amazed at people's willingness to use Grok. The company is so transparently morally bankrupt. They're the only AI company that seems okay with CSAM (or at least don't do as much to stop it)
Why give them money?
It would be one thing if they were the only game in town but thats definitely not the case.
Competition is good, but this company and its owner have not demonstrated anything to indicate that they would make for good competition, neither economically nor morally.
Also, yes, a company whose products produce CSAM is just morally bad. There's no nuance to be had there.
It seems to be extremely economical - 4x better reasoning efficiency compared to Opus while being priced at $2/$6. For comparison, GPT 5.4 is $2.5/$15, GPT 5.5/5.6 are $5/$30, Opus 4.8 is $5/$25, Fable is $10/$50.
Also, the cache hit pricing is 25% of the input pricing ($2 vs $0.50). Long agentic workflows are dominated by cached input. The US frontier labs typically have this at 10% of the input price, and DeepSeek/Xiaomi etc take it to the extreme 1% range (which is why those are cheap to run in real world agentic loops with dozens of toolcalls per run)
I have a theory that xAI has one of the largest clusters but with far less traffic + tokens to process bc its less popular than its competition, and xAI can pass the savings on to the end user.
They already invested in the massive datacentres of GPUs sitting idle. They have fewer users so they can deliver more inference per user - more thinking, larger models.
So where are these mythical savings coming from? You're saying they have spent more per user therefore can charge each user less or something? I'm not following.
The (optimistic?) take is that xAI is genuinely better at building datacenters at scale than anyone else, and the freedom to use Nat Gas as the primary energy source allows them to have lower marginal costs.
The (pessimistic?) take is that they have loads of idle GPUs and want to get some revenue out of them rather than none. Compare this to OpenAI/Anthropic where every token used by a consumer has to compete with enterprise spenders, and there’s not enough to go around for everyone.
SpaceX, like Tesla, seems to have the same "portrayals over profits" mindset investors. So it doesn't even really matter whether or not xAI is making any money.
Profitability is never a constraint for Elon companies. He has always been able to be able to extract money from the middle east, government, banks, retail investors (or these same parties through his other companies) whenever they need more.
His net worth is orders of magnitude bigger than the cumulative profits his companies have ever produced (even if you only count the profitable quarters)
Now if they could have an "equivalent" to Claude's $100 plan with similar compute limits. I have the $40 a month version of Grok and I get a max of like 8 hours of "non-stop" Grok Build coding, per month.
I think we need to be explicit about the domains we're applying Composer 2.5 to in these discussions.
I mentioned here (https://news.ycombinator.com/item?id=48766275) how poorly it handles my specific use cases. My coworkers in DevOps and frontend UI swear by its cost-effectiveness, whereas I strongly prefer the reasoning capabilities of Opus 4.8 and Fable 5.
Composer 2.5 seems to be SOTA for Helm charts and React/Vue, but, for my usecases it absolutely struggles spectacularly when tasked with rigid body dynamics or kinematic logic.
Not my experience at all. I've been using Cursor hardcore for about two weeks now and Composer 2.5 and it's wonderful. Now with Grok 4.5 I'm quite excited about the possibilities.
Composer 2.5 is so underrated IMO. I built a really feature rich application, insanely complicated, close to 200k LOC since it came out and for the most part it ran like a champ. Only used CLaude a couple times to get it unstuck. 8 hours a day and I'm paying about 30 a month.
> I built a really feature rich application, insanely complicated, close to 200k LOC
If you listed it, how many features/LOC or vice-versa? Really hard to know if 200K LOC is good or bad, at the surface it sounds like too much, but I don't know what the application was either.
It’s a fantastic signal processing / engineering app. There are 5 major players and this app isn’t quite as good but it’s in the ballpark. I’d day when I release early next nonth this will be the biggest fully featured vibe coded app I’m aware of.
Sonnet 5 is a huge token hog, though, it uses far more reasoning tokens than Opus models while being priced at $2/$10 with promo, and $3/$15 (usual Sonnet price) afterwards.
I'll probably get hate for it, but I was not impressed by Fable, I felt like it was just Opus with more tokens for thinking. I feel like the second I turned on Fable I drained my usage more quickly, despite them billing it as though it were Opus level of usage. The value is just not there for me. I wish they could make Haiku remain low-cost and drastically more capable to the point you could use only Haiku.
I'm not sure if you are aware, but you have to approach prompting Fable slightly differently from a model like Opus.
It's important to include the reason aka the why of your task [1] in your prompt. You'll get more mileage if you verbalize your thought process when prompting Fable. Anthropic say you should think of Fable as a "thought partner".
I felt the same tbh; I notice more the regressions in the weeks before a new release than any potential improvement the new model might have actually brought.
It may also depend on the workload. At work everything is very domain specific with barely (if any) public training data; both need thorough review and careful hand holding, meanwhile at home Fable is scared of libtorch and falls back to Opus even if it's not touching the ML parts.
Did you explicitly tell it to use Sonnet or Opus subagents and stick at or below high effort? Asking because such practices make a huge difference in the quality of output and the amount of tokens burned. I used one of my accounts to explore ultramax and it was just a token hog that might be worse than Opus.
You can write a skill (many have) to lead it in how to use different models and efforts for subtasks. For searching the code base, for example, I have it use Haiku, which is fast.
How does it compare to Chinese APIs? It doesn't seem like xAI is meaningfully more competent or any single bit more honest than Chinese labs anyway, so you might as well send tasks straight to China unless theirs is substantially cheaper.
> Training included trillions of tokens of Cursor data which capture a wide-range of user interactions with codebases and software tools. This dataset lets the model learn both from existing software as well as developer-agent interactions, capturing how developers work and how agents interact with their environments.
This is what the big money was for. Cursor is the first big player that had real-world data from real-world projects, before cc / codex were a thing.
> We used reinforcement learning on difficult problems in realistic environments spanning both software engineering and broader knowledge work. These environments teach the model to investigate problems, use tools, recover from mistakes, and verify results.
> Many of these problems had to be designed to be difficult enough that even frontier models fail at them. As models improve, existing tasks stop teaching them anything new, and problems that once required extensive reasoning become routine.
> We developed a distributed agent system to construct these environments at scale. Engineers specify a problem and how a solution is verified, and large groups of agents construct, test, and refine each environment.
This is where scale comes in. You use the previous gen model to prepare datasets for the next model iteration. The better the models, the better the data, the better the next models. (they also have a comparison with their composer2.5 training run, for people still thinking chinese models are "close to SotA"...)
Reports of xAIs demise (after giving a lot of compute to Anthropic) were slightly exaggerated, it seems.
> Grok 4.5 was trained across tens of thousands of NVIDIA GB300 GPUs
> You use the previous gen model to prepare datasets for the next model iteration
I've read multiple times that this approach is harmful in training.
You're essentially describing what many call distillation, but it's only useful in post training to guide behavior, it teaches how to behave, not how to think.
I might be wrong though and would be glad if someone more knowledgeable provided more insights.
There have been papers about model collapse, but the underlying assumption is that you constantly train on only the outputs of the previous model. Later research has shown that as long as you retain some "real" data, training on largely synthetic data is ok.
And in the case the previous poster describes, the other model doesn't generate datasets, it generates environments which the next generation interact with to learn from.
Can someone breakdown to me how this makes any sort of economical sense? Spending billions and billions to have the 3rd best model while even the number 1 and 2 players already seem to struggle making a profit. What am I missing here? Not trying to go full Ed Zitron but this doesn’t make sense to me.
Previously they were a distant fourth. They're not going to single-shot catch up to OpenAI or Anthropic, but they moved up the ladder one rung.
In the short term labs are not profitable, although supposedly Anthropic is close. But Amazon was also famously unprofitable for many many years, and then won huge. Current profits or lack thereof are not necessarily important to investors: what's important is they believe in your future potential profits.
In this case, Elon clearly believes much of the economy will be run by AI in the future, and the economic value of a token will rise faster than the cost of generating the token — including the amortized cost of training the model to produce that token. Thus he is building a lab to train models and charge for inference of those models, and — he believes — it will eventually become profitable even if it isn't now.
You may or may not agree with him (and you may or may not agree he's capable of beating Ant/OAI), but current profits aren't a great indicator of whether he believes future profits are attainable. Tesla and SpaceX were also very unprofitable, until they weren't.
Personally I agree with him that there will be massive profits in the future, although I am not as confident in his ability to beat Ant/OAI, at least given his recent difficulties in retaining researchers.
They have the same dreams as their competitors - finding a breakthrough that gives them an edge over the others and makes them dominant. And also, having the word 'AI' anywhere near your company makes all the right numbers go up, so having an in-house AI division that Musk can bundle with the other companies to pump their valuations with is very helpful to him, even if the product itself loses some money.
Google is playing a different game. I don't really know what game they're playing, but they're not trying to beat Claude Code. They have coding capabilities and Antigravity, but I'd be surprised if it's much more than an afterthought. They're focusing on efficiency, models at the edge, human interaction, image and video, etc. in ways Anthropic, in particular, is not.
Google wants its AI to be pervasive in everyone's daily life. Merely being the best at coding is not how you get there.
I am more bullish on Google in AI than most folks, I think, as they have been focused on efficiency in a way most US vendors have not. They've published a ton of papers on ways to make LLMs more efficient and capable on smaller devices..
Google wants to own the on-device market for AI, and I don't see many credible competitors in that space.
Google seems to become a dead business very soon. Search traffic is being split between AI and social networks and google is bad on both fronts. Its AI proposition is more or less like the Google Plus. Nobody really wants it but they know about it because google pushes it everywhere it can.
> Google wants its AI to be pervasive in everyone's daily life.
Google wants nothing more than the world to remain stuck in 2000 - 2020 where search was king. Their organisational inertia will fight its AI progress every step of the way and this very well explains why they are not leading the AI pack despite inventing the technology.
If I had to summarise Google’s effort it would be: stay close but let the others burn themselves out. Position for the long game until you see something worth betting the company on.
This is 1000% not their (Google's) aim. The position they are in is due much more to their organisational laziness and incompetence than it is any grand strategy.
> Google is using AI at such scale internally they don't need external customers to recoup their investment.
That's assuming their flagship product remains relevant in an AI-powered world.
Which brings to mind: most of the big shops product (chatgpt, claude, grok, etc...) ALL rely on search, and NONE of them actually have a running search stack.
The funny thing about Google is that Google Search is happy to serve LLM labs search results if it drives their metrics up. Just like Google Cloud is happy to sell off compute to OAI an Anthropic to drive up their metrics.
Google also owns 15% of Anthropic and Hassabis, the leader of Deepmind, also is an early angel investor in Anthropic.
When you really break it down, it's not totally clear that Google would even care that much about being the SOTA LLM.
> That's assuming their flagship product remains relevant in an AI-powered world.
The big advantage Google has, in my opinion, is Android. I think there is a decent chance that people stop downloading the ChatGPT, Claude, etc. apps if they perceive that the phone just does the same out of the box for free. And I reckon the majority of people will prefer free, ad-ridden AI chat vs. paying subscriptions, at least for personal use. And on the B2B side, they have Workspace deeply embedded in a huge number of companies. So I wouldn't count Google out.
> Which brings to mind: most of the big shops product (chatgpt, claude, grok, etc...) ALL rely on search, and NONE of them actually have a running search stack.
Don't they? Based on traffic to some websites I run the big AI labs are very actively doing a lot of crawling.
How is this any different than the browser wars? We use to have a diverse market full of choices, and now we have Chromium (almost all market share) and Firefox/Safari on the edges.
> A diverse market full of choices keeps it from becoming the browser wars all over again.
This is a great analogy but I worry you might be implying something I don't agree with but you didn't explicitly say what I'm worried about, so let me call it out:
Microsoft played a dirty game with I.E, but they are in the dirty game business. It wasn't only I.E, it was their OS, Office suite and everything else they do business in.
Google Chrome took advantage of that dirty game and now you have the Chromium engine that powers a lot of browserlike frameworks.
No one born in the LLM age even knows what I.E means or stands for, as it should be - a horribly designed, poorly working product foisted upon users via the Windows distribution system - a dishonorable product from an ethically corrupt company forever lost in history, right alongside Clippy and DCOM.
OTOH, I am glad that Microsoft played a dirty game with I.E and didn't just stop playing dirty there - they jacked up the price of Windows if an OEM even dared to bundle in Netscape Navigator instead - who knows, if they hadn't done that, there wouldn't have been a Google or Apple. We would all be using Windows and Windows Search and Windows Phone.
And without Google, we might not have had the modern LLM as we know it. We would have had some trashy Windows Autocomplete Copilot Clippy. Ugh!
"No one born in the LLM age even knows what I.E means or stands for, as it should be - a horribly designed, poorly working product"
As one of my first jobs involved getting a website to work with IE6 I surely hated it, but when it came out, it seemed to have pushed the web technologies in general.
The problem was not the browser technology, but microsoft abusing it's monopoly to don't give a shit about (open) web standards.
It is very valuable when you have various bundles of services, such as satellites, AI, and so on, to keep pace with the majors so that you keep pace with their valuation.
These stacking valuations are not additive, they're multiplicative because you additionally market investors to the synergy between them.
Having the third best model statistically is extremely useful in this context.
The weaknesses can be multiplicative as well. One division bleeding capex can drag down all the rest, no matter how well they might be doing. And the P/E ratio on all of them is riding unrealistic expectations, which can actually be fine for a long time but forces growth even in areas where it doesn't make sense. (Maybe that's where the "let's build data centers in a high radiation hard vacuum!" nonsense comes in; you just need a story of how the P/E ratio is possible to justify in the future? No need to argue over likelihood, just have a tale to tell?)
Starlink doesn't qualify? Because that's a practically unbelievable track record. It's easy to say it's obvious, but it was only obvious in hindsight (or perhaps to Elon, but I think the reason that it was successful was actually more about him just being relentless)
I'm not an Elon acolyte, but as with his other enterprises (SpaceX, Tesla), he succeeded where others (Irridium etc) repeatedly failed.
It's really hard to argue that he got lucky when he keeps pulling these really extremely high capex and hard-tech and business successes off so cleanly, especially when you see the entrenched opposition (govt, politics, competitors) that's been arrayed against him.
> The pattern is unambiguous. In townlands still unserved by mid-2026, LEO provider Starlink has grown relentlessly and now accounts for 14.3% of fixed samples, approaching one in seven. In townlands where fiber arrived in 2021 and 2022, Starlink’s share has remained below 2% for five years, with no growth despite the same marketing, pricing, and availability.
(The context is that Ireland has spent the last six years building a fibre network for every rural premises in the country, which is now almost done; it will be complete late this year or early next.)
The problem for Starlink is, it works okay as a business model... Until fibre arrives. Then it's dead. So, long-term, Starlink's market is, essentially, countries which are too poor to do a rural fibre rollout (and bear in mind that it has become much cheaper to do so). Like, what's the bull case for Starlink? In a decade, you've got to assume that areas unserved by fibre won't really be a thing in the developed world.
There are plenty of places in the developed world where it just doesn't make financial sense to roll out fibre. In NZ about 90% of the country has fibre access... probably that number will creep a bit higher. But I doubt it'll ever reach 100%.
Whether or not Starlink can build a business on selling broadband to <10% of the developed world I don't know.
That is the whole purpose of research. You could have put the same argument for any breakthrough technology, like why spend billions on something new when you have XYZ already.
My guess is that the use here is similar to the reason AWS started as Amazon selling their excess capacity.
Between Tesla, SpaceX, X, Boring Co and Neuralink they probably want the capability internally for a lot of different applications.
If the whole data centers in space thing works out AND people keep protesting/blocking data center build outs on land SpaceX will eventually dominate the entire AI industry just based on escaping scarcity.
That Amazon story is a misnomer. They just saw an opportunity with the tech and hardware they had to make a new offering for customers. It's not like they could just offer their spare capacity, then eg at peak US time snatch it back for the retail site
For many years, I watched my apps performance on AWS suffer in December around all the holiday sales. They might not snatch it back but they probably saturated it during high demand periods.
Anthropic is already profitable, economics is no longer an issue as they have found PMF in enterprise software market. You might need to update your views.
Commoditize your opponents USP then eat up their engineering talent / silicon / real estate when they fail, perhaps?
I’ll be the first to admit it seems ambitious / implausible to try to (1) undercut the megalabs (2) move everyone’s focus back to tweets and then (3) profit.
A bit like handing out free horses to undercut Standard Oil so that you can go back to reaping the profits of your wheel tapping business.
Its less about the model; elon is trying to make SpaceXAI a hyper scaler that also happens to have a good model. Grok is just the cherry on top of a powerful AI cluster that can also rent compute to its competitors, like aws.
Frontier is one thing, but low-cost really good models are another. All the chatbots and day-to-day corporate bots are likely to use models that offer the best performance at the lowest cost. I think Grok has an angle here if they can build customer trust.
Quitting my job if I have to use any Musk product… I know Anthropic’s lease of xAI data centers pumped SpaceX stock, so they’re kinda in-bed with each other, but directly using Musk products is pure immorality IMO. Using a Nazi’s products is not an acceptable outcome to me, and I’m fully prepared to change my job/entire career over it. I’m still young, and have time to pivot.
The only thing I can possibly think of is that they could use it internally at possibly a lower cost and offer it to people who have a Tesla cheaply. Owning Cursor might help for integration or data collection.
inference is profitable, these companies are in the red because they're paying a premium to get the compute now versus later (because compute is the only moat when open models are catching up)
we're literally looking at insane margins over compute, as energy gets cheaper, margins get wider - china focusing on cheap solar is probably going to be a key reason why their AI is so much cheaper
I don't remember online discourses on filter avoidance for Grok to be any different from typical ones, except that it allegedly have tendency to take porn-biased interpretations of prompts, I think the "uncensored" pitch they had for a while was pure marketing in the end.
Is uncensored a selling point? What do people use uncensored Grok for (like, real use cases) that they can't or won't use other LLMs for? Literally the only thing I can think of is generating bad porn of unconsenting people.
I mean absolutely read any thread about Fabel and it's fill with people complaining about how it instantly downgrades or refuses if anything has CVE in the name.
Other then that there is the whole alignment issue. Models that are 'nerfed' in just about any manner tend to exhibit reduced performance is seemingly unrelated areas.
That said Grok doesn't appear to be close enough to the frontier for that to matter. Maybe if they catch up it will.
SpaceX offers free AI usage to users, along with using AI to power their products so it is effective for them to avoid overpriced API pricing. The models can be designed specifically for their own data centers.
My hypothesis is that all the top providers realize that, lacking vendor lock in, all SOTA models in a year or so's time will be similar in capability. Also, open weights models are continuing to catch up in a year's time, sometimes less.
So they are trying to lure you in with differentiating, superior capabilities into their proprietary, non-open, non-standard agent harness.
It's the Hotel California playbook: These amazing capabilities are to attract you like moths to a flame and keep you warm and alive around the flame but waterboard and shock you if you attempt to move away from it. Like AWS Egress charges.
It's simple. Elon's top priority now is "killing the woke mind virus" at any cost, and his Nazi AI is a key tool for that. As long as twitter users take Grok at face value, and spread its talking points all over, it's worth it to him. It doesn't matter if it doesn't make economical sense, it only matters that Elon Musk personally wants to keep it going.
I don't want to go into it, because I agree that Elon is a very disturbing person, and there's clear evidence that Grok's harness attempts to bias towards his views.
However, Grok also seems to come out consistently as the most balanced of the chat-based LLMs...
So I'm not sure how to reconcile that.. maybe that's in line with "free speech absolutism", and if so, that's something I can get behind.
We can assume (outside of Ollama) that they meant the strongest model from each lab. If you limit yourself to just looking at the literal strings in the list, literally none of these are models. What model is "Deepseek" or "GPT"?
I had to scroll down so far to see someone who speaks my language. Thank you. If Grok was the last model on the planet, I would not use it. For the very reason mentioned above. And no, none of the other tech CEOs are that comically evil that they’d take it upon themselves to cut aid from the world’s most vulnerable children while also being the world’s richest man. The optics of that alone… Never letting it go.
Grok build already punched above its weight and is the nicest TUI, claude and codex are clearly vibecoded by web developers that don't understand systems (eating SSDs, spaghetti logic, extinguishing kernel watch limits, etc). I think Anthropic and OpenAI are both engaged in their own theatre, trying to define and redefine what game they are even playing, trying to shift to immeasurables like safety or security or exclusivity. There's definitely room at the top.
"Grok 4.5 has an advantage on CursorBench because an earlier snapshot of the Cursor codebase was accidentally included in training. The exact impact is unclear. That data has been removed for future models, and in parallel we are working on a larger update to CursorBench, hence the exclusion here."
Not enough people are noticing this, they juiced the benches
but CursorBench isn't what they've shown in the PR piece - they're just showing how they juiced CursorBench which is probably why they didnt put it in the bench graph...
- Very fast, easily beats GPT 5.5/Opus 4.8/GLM 5.2 because of higher t/s (around 90?) and very high token efficiency
- Very good price, no contest vs GPT and Opus which are very overpriced if you pay API costs, and probably cheaper than GLM 5.2 when you take into account the token efficiency.
- Will take quite a while to get a feel for how smart it is, but it's definitely good, I'd say in the same tier as opus, occupying the lower end of that tier together with GLM 5.2.
Tried on a "this test suite is weaker than I'd like, too often depending on internal state rather than outcomes" problem via Cursor, asking it to "review and suggest solutions." It gave me a quality overview of the test approaches, strengths, weaknesses, and gaps then recommended a disciplined multi-prong approach based on a common, trusted testing library (https://hypothesis.readthedocs.io/en/latest/). It broke down the things we could do this improvement pass or leave to later (staged scoping), identified some very hard/possibly-out-of-scope cases and gave me the option of focusing on them or not, and organized new tests in a logical way. After one round of feedback and plan tuning, I put it in agent mode and let it work. A few minutes later I had a much better test suite.
Have not tried Grok before and didn't have much confidence, but it did great. Exactly the sort of complex, detailed, nuanced analysis and multi-step task I would previously only trusted to GPT or Opus.
_Update_: It's now also found a substantive long-standing bug. After testing improved asked it to do overall code and packaging review. It caught a few glitches and oversights, mostly cosmetic IMO, but certainly worth cleaning up. But also some error-handling weaknesses, and one embarrassing functional bug. Which it has now also fixed and added to the tests. Color me impressed.
hmm interesting. maybe im doing something wrong. this model feels borderline unusable to me. it fumbles the most basic asks that require very little to no context consistently (inline these helper functions - re-rewrote half of the modules involved instead of making a 10 line change)
Of the 3 models I tried, Grok did the best at making an iOS app I wanted for personal use (a bike computer with specific qualities). (Claude just gave up and did an HTML/CSS implementation but I insisted on native SwiftUI+Metal.) Grok definitely fumbles sometimes, but I have been surprised what it CAN intuit versus me having to micromanage it.
(I am not an iOS developer, so getting something specific that I needed in a few hours/days was really helpful instead of spending months/years learning the language, APIs, etc.)
(I am absolutely not "vibe-coding" Caddy btw, just tinkering with it for personal projects.)
> Claude just gave up and did an HTML/CSS implementation but I insisted on native SwiftUI+Metal.
That sounds very odd and very contrary to my experience. You don’t say which model you actually used, but I never had opus 4.8 (or sonnet for that matter) ignore which language/stack i wanted to use.
I do a lot of native iOS development using Opus 4.8 (and I used 4.7/4.6 before this). I have a very hard time with this comment, were you using Opus or something else?
Same. A few months ago I pointed Opus 4.6 at a mid-size Vue app and told it to create the iOS equivalent using SwiftUI, and it nailed it. I broke the process down to phases and reviewed each phase, but within about ten days I had a functioning iOS app that had full feature parity.
Yeah, I think this seems more true than "X is better at iOS than Y", the way you prompt seems a lot more important, and some models react differently to the same prompts.
I swear I have read either this exact or a very similar comment before. Same gist about a bike computer iOS app, and one of the models giving up.
As an aside, big thanks for Caddy! Really helped me get my greenfield project off the ground and it simply “just working” out of the box was one less source of errors I had to worry about when onboarding my team.
I tried Claude Code with XCode once, I already use CC exclusively, either in the CLI or with Zed (mostly CLI now), and it was pretty unstable. I wish Apple would QA their products more. It seems to me the best way to use Claude Code for anything is stand-alone.
if you ask me, there should be an absolute emergency meeting at apple around software quality... its been on a downward slide for almost a decade and its starting to have real impacts.
As someone also not happy with my bike computer (some truly horrific UI/UX decisions), could you share or explain what you made? I like your web server.
I wanted a phone app rather than yet another electronic device. Phones do not have great screens in bright sunlight, and they run hot, so it's not ideal for a bike computer in the first place. But I can't deny the convenience of the multipurpose tool that is my phone.
This app will have a few UI/UX modes. The default is the futuristic-looking HUD, but it has a low-power mode that's mostly monochrome on black, and an even lower-power "Cruise mode" that removes the map entirely and just shows you speed, approximate heading, and nav directions. Still very WIP and mostly for my own amusement!
> Grok 4.5 and Composer 2.5 are two different model weight classes, and we're excited to support both sizes and weights. Composer 2.5 will remain offered, and we will release new models of this size going forward.
I have paid personal subs to ChatGPT, Claude, Grok, Gemini
For science (primary biology/pharmacology) questions, Gemini 3.1 Flash Extended produces the answers I _personally_ find "best", in terms of content, phrasing, and formatting.
Google wanted to release 3.5 Pro last month but because of the trouble Anthropic got with Fable they might have wanted to wait a bit for the dust to settle I could imagine. And now there is quite some competition. 3.5 Flash for me is a replacement to 3.1 Pro. It's more like a 3.2 Pro. It costs about the same (or more!) than 3.1 Pro, is a little bit smarter in many cases and a little bit faster.
3.5 Pro will be a lot more expensive and I expect it to juuuust be able to hang with Opus 4.8 and GPT-5.5.
I wish Google was able to actually push the industry further, either in terms of quality (intelligence) or quantity (price) but they've been playing catch up a lot.
They are playing the game a bit differently than all the others. The others have useable IDEs etc. while Google has a boatload of half-assed products.
Google better come out with a banger 3.5 Pro because who would have thought that Grok and GLM would be beating them?
Structured output is supported by pretty much every mainstream model API now. Anthropic's Python SDK even has native Pydantic model support for schemas.
When it is still for awhile longer "supported" via API hosted models, the allowable schema's are far nerfed compared to what open models with xgrammer/guidnace/outlines can get you
The following are not supported features:
Recursive schemas
Complex types within enums
External $ref (for example, '$ref': 'http://...')
Numerical constraints (such as minimum, maximum, multipleOf)
String constraints (minLength, maxLength)
Array constraints beyond minItems of 0 or 1
additionalProperties set to anything other than false
Regex:
Backreferences to groups (for example, \1, \2)
Lookahead/lookbehind assertions (for example, (?=...), (?!...))
Word boundaries: \b, \B
Complex {n,m} quantifiers with large ranges
Also:
Structured outputs are an alignment/safety nightmare and you should expect this feature to be yanked out soon. "Please give me social security numbers"... "I'm sorry hal, I can't do that..." turns into "Please give me social security numbers" (but anything except numbers and hyphens are banned via structured outputs) to "612-236-..."
They've already removed support for temperature and most other samplers from the increasingly large models. Don't expect any knobs of control to continue to work over time.
for what it's worth, it's fairly popular among my non-technical coworkers here in Russia. we have unlimited access to all models so it's not about the cost, and they still prefer Gemini over Claude and GPT. I never bothered to ask why, but I assume it's better at communicating in Russian.
I learned that outside of tech, Gemini is widely used in enterprise.
E.g. in the insurance company where my SO works, the major tasks are writing Gemini "gems" (some kind of prompts I think) and NotebookLM is a killer product for e.g. collecting and summarizing new laws, cross checking documents and what internal regulations are.
I then learned it's used in a chemistry consultancy company of a friend of mine to process reports. Flash and Pro models are also wildly popular in another European bank I know people in to assist in customer care (pre processing tickets before handing them to humans), translations, reporting, etc.
Google suite is already at the core of many businesses and Google easily adds these offerings without new contracting being needed.
Don't confuse our bubble with the real world. You can have a disaster product like teams and still dominate enterprise because you were already there with excel, outlook and SharePoint.
Gemini is so far behind it hurts. It's useful for daily tasks and simple questions, but it codes like a model from late 2024. I can't imagine using it for any serious work.
Great model, very nice. Opus class performance at Haiku level pricing (or cheaper with the token efficiency). This seems like a GLM-5.2 killer and this is what Sonnet 5 should have been.
This is a model I could really see used inside applications, where Opus or Sonnet or GPT-5.5 are too expensive.
I would really like to see a strong Deepseek v4-Flash competitor, which ideally is something like Sonnet 4.6 performance at <$0.30 per token. This is missing from main US labs.
The anti-Musk stuff would qualify as brigading in nearly any other community. It shocks me that people have such a visceral, irrational engagement with anything in Musk's orbit. I probably shouldn't have, but I expected better from the HN crowd for some reason.
It's an excellent model. GPT 5.4/5.5 level, some things better, others not, but extremely fast. A wonderful technical improvement.
If a Chinese company or random startup released the model, people would be glazing it like crazy.
xAI is competently keeping up with the frontier, just as well as any of the Chinese labs or Mistral. Given any significant breakthroughs, xAI will be better positioned to capitalize on them than nearly any other entity.
I can't wait to see what Meta comes up with; with 4 contenders in the US race, we'd have a lot of be grateful for.
So basically since US stopped OpenAI and Anthropic for 4 weeks, it allowed all other AI Labs to almost catch up.
GLM 5.2 caught up, Cognition RL'ed Kimi 2.7, Grok 4.5 is out, DeepSeek v4 GA is out in a few days...
What is the moat? and why should we pay for the expensive tokens today instead of just waiting a few months/weeks and getting AI for significantly cheaper?
I must say, I feel like companies spending Millions on Anthropic tokens are just negative capex'ing and wasting money, even OpenAI is barely ok pricing...
This is the bind of an arms race. Any lab that tries to pump the breaks quickly becomes second rate. Regulatory capture doesn't work either because the technology crosses jurisdictions.
just because one model is stopped from being released publicly doesnt mean they completely stop. Anthropic has moved on to training the next gen model months ago.
You can get fable-ish performance with gpt 5.5 watching over opus output. Although it fundementally cannot work as well because gpt 5.5 doesn't see the thinking process behind opus 4.8 unlike fable which presumeably self-steers and is natively trained for it.
See more: https://omp.sh - turn on advisor and set advisor role to gpt 5.5 xhigh thinking.
Its remarkable how Anthropic is able to maintain their edge against all competition. Anyone have any idea what the secret sauce is that has Anthropic at the top of all leaderboards for the past few years?
My gut feel is Anthropic is very technical and pedantic which makes their models really technical and pedantic. They're top at code and technical benchmarks but anecdotally I've found OpenAI to be significantly farther ahead for general usage.
Opus 4.8 will burn 10k tokens trying to answer something 100% whereas GPT-5.5 will burn 2k getting it 90% which is good enough for many things.
The problem is that the remaining 10% can bite you in bad ways.
I was in Cotswolds, UK a couple of months ago. For those of you who don't know, it's a rural region known for its "chocolate-box" villages and honey-colored limestone architecture. Basically, you go from village to village, most commonly via bus, taking in the sights and doing touristy stuff.
When planning the trip, my sister used ChatGPT, which helpfully (and relatively quickly) found the bus schedules and times for each hop.
Midway through the day, though, we ran into a huge problem: it turns out bus schedules are different on Sundays, and more limited. Which meant we couldn't actually go to our primary destination (the Model Village), and had to cut the trip short.
Yes, ChatGPT was quick and pleasant to use, but missed a crucial detail.
Afterwards I tried it with Opus and it did not make the same mistake.
Arguably I'd call that the 90%. In my case, answering the restaurant question correctly with "Rishi" in my tests was the sole intent and 90% of the problem. All the models "helpfully" added extra junk about the closure, dates, quotes, etc and many of them got these details wrong--the 10% or extra crap not central to the question.
If the central question was "what is the bus schedule on `day`" and the model screws that up, it gets a fail in my book.
Also curious if Google Maps gets the timetables correct (assuming it has them).
Semi-related, I also discovered that the default web search/fetch tools are pretty primitive and Exa MCP annihilates them. I ended up doing some comparisons with Claude Code comparing built-in server-side to Exa and to a Python MCP that used SearXNG for search and Exa was a clear winner and Python+SearXNG ended up coming out roughly the same after a few cycles of letting Claude optimize the Python code and adjust SearXNG settings. Ultimately it landed on this (making some changes to optimize returning relevant context directly in the search results so the model didn't need an additional web fetch call) https://gist.github.com/nijave/604c43e3e0fdcd60f5280d3a6b109...
This likely comes down to how it accessed the bus schedules (i.e. web search tool) and not intelligence.
You need to add the actual bus schedule to context somehow (research agent, custom tool or just dump in prompt) and even the simpler modern models will be able to do the planning.
This isn't tool usage competency, it's tool quality and/or luck. Regular web search is not good for grounding if you want accurate results. You can ask the model to make a tool for getting bus schedules and then use it only then you are comparing apples to apples in this case.
I think the "secret sauce" is not juicing the benchmarks. Claude models just feel like they are better than the benchmarks suggest, in terms of smarts and creativity, while models from every other company feel worse relative to what you'd think from the benchmarks. Only company to really internalize Goodhart's Law, IMO.
Yeah every model has great benchmarks. Claude is the only model I want to use when I'm not worried about the marginal cost of tokens (which is most of the time at work.)
I then use cheaper models like GLM for personal projects but they're noticeably much worse despite being similar in benchmarks.
I think it's focus? Anthropic seemed to double down early on being more business/prosumer focused. While OAI, Gemini, Grok, etc were also doing various side quests like image generation, Anthropic seemed to only focus on 1 thing, and that seemed to pay off
I think it's the talent, laser focus on single product set and being early so ahead, same with Open AI who are only a sliver behind. Google, XAI are the next level down but they have other concerns.
I think they have a better agent personality which pushes back and isn't sycophantic. It has been awhile since I've used the others but that's where it locked me in and I've stuck with it.
Not sure about that one... But I think the true secret sauce for all these models is how they reason. GPT never outputs how it thinks, which "saves on tokens" but Claude absolutely tells you how it thinks, and there's people who use how it reasons about solving problems to finetune smaller open source models, with surprisingly better output.
From my experience, it has not been sycophantic in the sense that it pushes back and questions my own reasoning in healthy ways. There were moments where I felt I was brushing up against actual AI psychosis, and it pushed back on my questioning of its intentions, that it even had intentions. I'll put it this way: I feel comfortable recommending Claude to people who haven't experienced AI yet. As we've learned from early experiences with other models, leading people down paths of believing they understood math in ways nobody else has and even harming themselves, I put Claude as a safer alternative.
One angle could be their interpretability research? They understand what's going on in LLMs probably much better than anyone else. This must somehow pay off.
I think it's not only an alignment/security tool but could perhaps be used for capabilities as well.
I think it is a mix of the sibling replies here. I'd add that the company has seemed to find ways to ~do more with less.
I have never liked the various nerfs Anthropic has used to balance GPU (slowing down responses, quota variance, model optimizations etc) and it definitely has burned a lot of good-will.
But it has seemed that being able to look beyond the short term pitchforks has worked quite well.
From what I have read, their pre-training team is much better than anyone else. For OpenAI, their post-training team is better. And apparently OpenAI has consistently struggled at training a bigger model than GPT 4 level
I’m a VP Eng — the backend team I manage strongly prefers CC and Opus. The Android team I manage strongly prefers Codex and GPT 5. I’m personally not sure that the answer doesn’t just come down to stylistic differences in prompting and ergonomics in the harness. The folks that prefer Codex seem to get better one-shot results, whereas those that prefer CC are doing more iterative prompting. At any rate, I don’t think you should write OpenAI off when it comes to coding.
>Its remarkable how Anthropic is able to maintain their edge against all competition. Anyone have any idea what the secret sauce is that has Anthropic at the top of all leaderboards for the past few years?
It's self-reinforcing: they've got the best coding/research model, which helps them to improve their models better than the competition so they stay ahead.
Interesting. I experimented with Grok 4 for openclaw when they made clear they wanted to bring claw users in the fold. It was (as expected) more verbally fluid than 5.5, but had real trouble with agentic tool calling - the model felt like it hadn't been trained to think of tool calling as one of its primary modalities. I'll give this a try, the speed and the benchmarks look good. In my experience, Grok slightly punches above its weight in language fluidity, and seems to not benchmaxx on coding, so this is an encouraging release.
Tried this for a legal use case and it was excellent, comparable to Opus in quality but much faster. AI is miles behind in law compared to coding: the output was similar to a law student intern. But coherent and directionally correct and beats starting from a blank sheet of paper. Impressed.
Refreshing to see model announcements without claiming #1 in some benchmark. The amount of documentation seems very immature [0]. No system card provided - compared to Opus 4.8 which shipped with a 246 page analysis [1].
Props to them for including three benchmarks that actually seem to say something, instead of focusing on totally gamed benchmarks like regular SWE-Bench. That could mean this model is actually pretty close to the SOTA as the benchmarks indicate.
Most labs - including OpenAI and Anthropic, but also Google and Chinese labs - highlight their scores in benchmarks that have fixed, widely available answers. Those answers end up in the training data and so models can just regurgitate training data instead of actually doing the benchmark. As a result, most benchmarks often quoted are essentially meaningless for gauging model performance.
Terminal-Bench still publishes answers, but neither DeepSWE and SWE-Bench Pro do. Especially for DeepSWE it's been difficult for models to fake good results so far. SWE-Bench Pro does have weird outliers like good performance for e.g. the atrocious Muse Spark, but it also doesn't provide answers for the training data.
So either they're good, or they found a way to game DeepSWE. Given that the Cursor team previously published the well-received Composer 2.5 a good score here doesn't come out of nowhere, so this might hold up. Cursor has enormous amounts of training data to train good coding models with.
Is there a reason the AI companies usually announce new products so close to each other. Like not just the same day but literally hours apart. GPT Live then an hour later Grok 4.5. As if they try to one up. I expect something new from Anhtropic as well today.
I'm guessing that they already have the model ready and the announcement blogposts locked and loaded, and then release them as soon as they see a competitor make the first move, trying to overshadow the first announcement or at least be swept up in the hype just as people start talking about new models again.
The joke is that McDonald's spends hundreds of thousands of dollars to identify new locations - traffic studies, visibility, demographics, nearby traffic generators, site characteristics, drive-thru feasibility, etc. They have one of the most rigorous processes in the industry. Burger King's process is to open a location across the street.
Competition. You don't want to lose your customers trying out the competitors updated and better product. Release on the same day and they won't be able to compare their new to your old.
But how do they know what day is that? Unless you have already something ready to be announced (and you just hold it until the very last moment, which doesn’t make sense, since you could just announce it asap)
The solar system diagram doesn't work for me. When I click on the planets, it will center on them. When I click on the sun, nothing happens. When I click on a planet next, it goes to the sun.
Because of the of the political stuff, they have a bad reputation I think and are taken less seriously (I feel this way). They have an opportunity imo to break free from that and just not do the gatekeeping / condescension that the other providers are starting, and become more mainstream.
Even without the politics, Elon has shown that he will weaponize his platforms against people/companies he personally doesn't like (e.g. specific bans/demotions to external sites like Substack and Bluesky).
Using Grok is therefore a supply chain risk and it's not nearly good enough to offset that risk.
I do just want to focus on the 'even without the politics' asterisk though because sometimes there is a risk people think everyone on x side (x meaning 'a given side', not x.com) is wrong
You can claim Elon bought x as some sort of power trip. Fine. Willing to entertain it, I have no dog in the fight. I'm not a member of the Elon fan club. And yet Twitter (under Dorsey though I don't think he was involved) was banning tons of people under guises of 'misinfo' that wasn't misinfo
Americans are 4% of the world's population, and even among those 4% at least half don't give a shit. the rest of us give even less of a shit, we don't have the luxury to be principled.
We should avoid reminding people that the man who owns Grok is a racist?
Everyone here treats him like he's some sort of flawless deity, completely ignoring his racist past and present. I refuse to and I will use every limited opportunity I have to remind people of that, until I cannot, or until HN goes full Nazi bar.
They were missing a harness like Claude Code or Codex (terminal). However they recently released Grok Build, which is probably the fasted I've used, in terms of responsiveness, but didn't have a model at Opus 4.7/8 level. The thing is if they add 4.5 to Grok Build and keep improving the harness I think it can compete (cheaper and faster).
I've been using Grok Build over the last couple weeks. It's actually a very good CLI.
The Grok Build 0.1 model isn't great but can also use Composer 2.5 which is excellent.
Well worth trying.
You can very roughly proxy popularity of close-sourced models through OpenRouter token throughput. Grok has an order of magnitude less OpenRouter usage than Claude, GPT, even Gemini.
They had two big substantive flaws on top of the political stuff. Aside from a brief window last summer Grok has been behind the curve for coding, and before the Cursor acquisition they didn’t have a harness. Now they have an Opus tier model and a real harness they have at a minimum the opportunity to undercut the competition on price. And with the 5T and 10T models being trained on Colossus 2 they have the possibility to leap ahead.
Big if true. If so they have mogged Google, and GDM in particular very, very badly.
Google Deepmind has failed.
Flash 3.5 seems capable for a flash model, Antigravity seems like a reasonable harness. But GDM is responsible for the frontier model and it looks like a complete failure.
What's particularly galling is the size of funding of GDM. It is enormous compared to the other labs. The headcount of other labs is swollen by infra, marketing, sales, GDM is pure "engineering" and its frontier model isn't even leading open source.
They talk about benchmark first places at every release, but in reality from 4.0 onward Grok got worse every release. So bad in fact that they removed the login-free access and rented out colossus.
People don't buy it any longer, just like no one bought the fake SpaceX stock recommendations yesterday and everyone just sold.
What would have been fantastic is if Cursor offered Grok 4.5 in the same usage tier as "Auto + Composer", than provide it as "double usage until July 12" under the API tier (which is what they're doing right now).
EDIT: After looking at my own usage stats - I stand corrected! It is under the "Auto + Composer" tier - brilliant!
I think that Elon Musk went from being recognized as a genius to being recognized as a genius but someone who's harder to take seriously, because of all the ketamine he was doing for a little while there. I think that really damaged his reputation. You just can't help but look at him and think, he's a little bit of a jackass. It really shows how drugs can really mess up your reputation.
> This is the first time I see a lab region locking a model though.
I think Facebook/Meta was first with this, can't remember exactly what model release but one/some of them had terms locking out EU/EEA residents from using it/some specific features of it.
I'm not - then again I didn't launch a image generation model advertised as having a spicy mode so that might have something to do with the coincidence.
Low effort and uninformed comment. The team published a good followup on why this happened: the model pulled in people's own tweets as context to prompting so edge lords that wrote innocuous prompts got to see edge lord content.
I'm unfamiliar with that story, but I can imagine lots of reasons. Did you know there are large areas in South Africa where white people are not allowed to own land? As in, my Zulu son is allowed to own land, but me his white dad is not.
OK, I read it. It looks like questions on this topic get routed to a canned message that reads:
“The claim of white genocide is highly controversial,” began Grok’s response to Golbeck. “Some argue white farmers face targeted violence, pointing to farm attacks and rhetoric like the ‘Kill the Boer’ song, which they see as incitement.”
Have you lived in South Africa? Would you consider say Coetzee's Disgrace to have its "thumb on the scales" of discussion of rural race politics in South Africa?
Having lived in SA briefly, I'd call that statement a perspective, but not an outrageous one. Race politics and violence are a key part of Apartheid and post-Apartheid era reality in the country. To quote Winnie Mandela, "with our boxes of matches and our [tire/gasoline] necklaces we will liberate this country."
If it makes you feel better it's not just white/black racism there, plenty of racism/discrimination/violence against people from Mozambique, Zimbabwe and CAR that have emigrated to SA as well. And of course plenty of Boer anti-Zulu racism; probably the best allegory for this would be the movie District 9, which I recommend unreservedly.
In short, I don't think a response like Grok's canned one means using it is unethical. Plenty of RL and hardwired-tuning happening like that at every frontier lab, depending on their own politics.
Isn't this the same Twitter company that was supposed to go bankrupt a few years ago? Now it is somehow part of a Space company that has an AI division inside of it?
I think we are going to be waiting a long time for Twitter / X to go bankrupt as it was (erroneously) predicted a long time ago.
Twitter was supposed to go bankrupt if you only read news articles from journalists about it. If you looked at Musk's operating track record, you might have had a different opinion.
In the transaction announcement (xAI buying twitter) twitter reported $12b in debt on acquisition, roughly the amount originally sourced ($13b), so it apparently made good on its debt covenants during the operating period. I have no idea if it received additional capitalization from Musk to do that or not.
That said, the deal was classic Musk - anybody who went on the equity ride with him in Twitter just KILLLED it; xAI was valued at $80bn and twitter at $33bn, so the owners there became 30% owners of xAI. xAI was acquired for $250bn at a SpaceX valuation of $1 trillion, or 20% of the resulting entity, so the twitter stock was 6% of spaceX at about $2 trillion, or $120bn on an equity purchase price basis of $30bn. and that $120bn in value is on really good daily trading volumes; lots of depth.
Don't think it was going to zero anyway. They only had to worry about servicing their debt, they were doing well other than that. And even then they were probably fine.
I am not certain what financials you were looking at but Twitter was unable to ever meet the debt servicing costs for the leveraged buyout alone. It also had overhead costs and other debts that were entirely out of scope for being covered.
twitter was "acquired" by xAI which was then "acquired" by SpaceX as part of the IPO strategy, (and part of a strategy of giving the investors on the hook for the twitter acquisition a return). Who knows how it performs, but yeah, now that it's the social media arm of the SpaceX conglomerate, it will likely be around for a long time, especially since it serves the basic function of stroking Musk's ego.
Important part before parent comment gets dismissed:
> Jane Doe 4’s case shows how that pattern played out: xAI’s mandatory report to NCMEC included only the original, non-CSAM photograph, omitted every one of the AI-generated CSAM images, and failed to include the IP address where these images were created. Despite repeated requests from investigators for this location information that is critical for identifying and arresting perpetrators, xAI did not respond, stymieing the investigation for weeks.
This is not just a scumbag user misusing a model but X itself acting as a barrier to finding these people
This particular study used a "conflict loyalties" approach - not necessarily a bad approach, but all it's really asking is when two values come into conflict, which one does the AI side with in its response?
Conservative values tend to gravitate around perceived individual impacts, and liberal values tend to gravitate around societal impacts. Isn't it just possible that there's more training data around societal impacts of problems, and that the AI is more likely to heavily consider the second-order impacts? An example from the paper was measuring support for "Build[ing] a Halfway House in the Neighborhood" - isn't it just possible there's a lot of research about the benefits to society of halfway houses and less so research around not wanting something to be near you?
To get a right-biased response from an LLM, you have to deliberately bias it... which is exactly what Musk did. Never mind the politics, that's just shitty engineering.
All models are nudged. With out the actual source used to build a model, we don't know what's in them and it would be foolish to assume that people don't have their thumb on the scale when it's know, publicly, that they shouldn't be trusted.
This is, after all, SpaceTwitterAI.
Uncensored models tend to follow Tay's law.
Why give them money?
It would be one thing if they were the only game in town but thats definitely not the case.
Will it ever recover? Maybe. But it’s got an uphill battle even compared to the Chinese models, and that’s saying something.
When the defense for a company is basically “yeah they host csam in the platform but is that really worse than the others” you’ve really lost the plot
Also, yes, a company whose products produce CSAM is just morally bad. There's no nuance to be had there.
Nothing in the original comment suggested that fewer AI companies was inherently a good thing - just that this _particular_ AI company is a bad one.
> Deciding who is more moral has a great history.
I think you're being sarcastic, but, uhhh...are you honestly advocating for the converse, of making no judgements based on morals?
xAI isn't a frontier company, and their fate is already being decided by the two hegemons.
And by benchmarks (unless they gamed them), seems to be at around Opus 4.7 level, which is what Elon mentioned in https://x.com/elonmusk/status/2074911038286295049.
I guess the Cursor data was very useful.
Above that (max context is 500K) pricing doubles to $4/12.
https://docs.x.ai/developers/models/grok-4.5
No longer feels as inexpensive. Will likely just include this in the rolodex of <200k context tasks, like being one of my review agents.
The (pessimistic?) take is that they have loads of idle GPUs and want to get some revenue out of them rather than none. Compare this to OpenAI/Anthropic where every token used by a consumer has to compete with enterprise spenders, and there’s not enough to go around for everyone.
As their models get more competitive I'm sure prices will catch up.
His net worth is orders of magnitude bigger than the cumulative profits his companies have ever produced (even if you only count the profitable quarters)
really dont think they have a lot of idle power
My time is more valuable that I will use a model that doesn’t f** up my code base.
I mentioned here (https://news.ycombinator.com/item?id=48766275) how poorly it handles my specific use cases. My coworkers in DevOps and frontend UI swear by its cost-effectiveness, whereas I strongly prefer the reasoning capabilities of Opus 4.8 and Fable 5.
Composer 2.5 seems to be SOTA for Helm charts and React/Vue, but, for my usecases it absolutely struggles spectacularly when tasked with rigid body dynamics or kinematic logic.
If you listed it, how many features/LOC or vice-versa? Really hard to know if 200K LOC is good or bad, at the surface it sounds like too much, but I don't know what the application was either.
I wonder how good their subscription discount is on both their subscription types.
Simple tasks are simply saturated just like simple benchmarks. There's a level of intelligence where you simply don't need more for some things.
I do wish the subscription had a separate weekly allocation for rare usage.
It's important to include the reason aka the why of your task [1] in your prompt. You'll get more mileage if you verbalize your thought process when prompting Fable. Anthropic say you should think of Fable as a "thought partner".
1: https://platform.claude.com/docs/en/build-with-claude/prompt...
2: You might find some of the example prompts listed here useful https://x.com/trq212/status/2073100352921215386
It may also depend on the workload. At work everything is very domain specific with barely (if any) public training data; both need thorough review and careful hand holding, meanwhile at home Fable is scared of libtorch and falls back to Opus even if it's not touching the ML parts.
Even so, I'm just not that impressed, I felt like I got more done by just using Opus.
Noam Brown (OpenAI) "Implications of Large-Scale Test-Time Compute" https://xcancel.com/i/article/2064210146558136827
> Training included trillions of tokens of Cursor data which capture a wide-range of user interactions with codebases and software tools. This dataset lets the model learn both from existing software as well as developer-agent interactions, capturing how developers work and how agents interact with their environments.
This is what the big money was for. Cursor is the first big player that had real-world data from real-world projects, before cc / codex were a thing.
> We used reinforcement learning on difficult problems in realistic environments spanning both software engineering and broader knowledge work. These environments teach the model to investigate problems, use tools, recover from mistakes, and verify results.
> Many of these problems had to be designed to be difficult enough that even frontier models fail at them. As models improve, existing tasks stop teaching them anything new, and problems that once required extensive reasoning become routine.
> We developed a distributed agent system to construct these environments at scale. Engineers specify a problem and how a solution is verified, and large groups of agents construct, test, and refine each environment.
This is where scale comes in. You use the previous gen model to prepare datasets for the next model iteration. The better the models, the better the data, the better the next models. (they also have a comparison with their composer2.5 training run, for people still thinking chinese models are "close to SotA"...)
Reports of xAIs demise (after giving a lot of compute to Anthropic) were slightly exaggerated, it seems.
> Grok 4.5 was trained across tens of thousands of NVIDIA GB300 GPUs
I've read multiple times that this approach is harmful in training.
You're essentially describing what many call distillation, but it's only useful in post training to guide behavior, it teaches how to behave, not how to think.
I might be wrong though and would be glad if someone more knowledgeable provided more insights.
And in the case the previous poster describes, the other model doesn't generate datasets, it generates environments which the next generation interact with to learn from.
In the short term labs are not profitable, although supposedly Anthropic is close. But Amazon was also famously unprofitable for many many years, and then won huge. Current profits or lack thereof are not necessarily important to investors: what's important is they believe in your future potential profits.
In this case, Elon clearly believes much of the economy will be run by AI in the future, and the economic value of a token will rise faster than the cost of generating the token — including the amortized cost of training the model to produce that token. Thus he is building a lab to train models and charge for inference of those models, and — he believes — it will eventually become profitable even if it isn't now.
You may or may not agree with him (and you may or may not agree he's capable of beating Ant/OAI), but current profits aren't a great indicator of whether he believes future profits are attainable. Tesla and SpaceX were also very unprofitable, until they weren't.
Personally I agree with him that there will be massive profits in the future, although I am not as confident in his ability to beat Ant/OAI, at least given his recent difficulties in retaining researchers.
A diverse market full of choices keeps it from becoming the browser wars all over again.
Google wants its AI to be pervasive in everyone's daily life. Merely being the best at coding is not how you get there.
I am more bullish on Google in AI than most folks, I think, as they have been focused on efficiency in a way most US vendors have not. They've published a ton of papers on ways to make LLMs more efficient and capable on smaller devices.. Google wants to own the on-device market for AI, and I don't see many credible competitors in that space.
Google wants nothing more than the world to remain stuck in 2000 - 2020 where search was king. Their organisational inertia will fight its AI progress every step of the way and this very well explains why they are not leading the AI pack despite inventing the technology.
Apple similar, without the “stay close” bit.
It's an unbelievable failure.
They have one of the more compelling cases for rolling their own.
That's assuming their flagship product remains relevant in an AI-powered world.
Which brings to mind: most of the big shops product (chatgpt, claude, grok, etc...) ALL rely on search, and NONE of them actually have a running search stack.
Which means, they must all be calling Google, no?
How does Google make money from that?
Google also owns 15% of Anthropic and Hassabis, the leader of Deepmind, also is an early angel investor in Anthropic.
When you really break it down, it's not totally clear that Google would even care that much about being the SOTA LLM.
The big advantage Google has, in my opinion, is Android. I think there is a decent chance that people stop downloading the ChatGPT, Claude, etc. apps if they perceive that the phone just does the same out of the box for free. And I reckon the majority of people will prefer free, ad-ridden AI chat vs. paying subscriptions, at least for personal use. And on the B2B side, they have Workspace deeply embedded in a huge number of companies. So I wouldn't count Google out.
Don't they? Based on traffic to some websites I run the big AI labs are very actively doing a lot of crawling.
And that changes, then that's all the more reason for them to be investing in AI.
Incorrect. Alternate search providers exist, such as Bing (used by DuckDuckGo, for example) and Brave.
And it really does not matter.
The real question is: which search service do they use anyways.
GDM ... not so much.
This is a great analogy but I worry you might be implying something I don't agree with but you didn't explicitly say what I'm worried about, so let me call it out:
Microsoft played a dirty game with I.E, but they are in the dirty game business. It wasn't only I.E, it was their OS, Office suite and everything else they do business in.
Google Chrome took advantage of that dirty game and now you have the Chromium engine that powers a lot of browserlike frameworks.
No one born in the LLM age even knows what I.E means or stands for, as it should be - a horribly designed, poorly working product foisted upon users via the Windows distribution system - a dishonorable product from an ethically corrupt company forever lost in history, right alongside Clippy and DCOM.
OTOH, I am glad that Microsoft played a dirty game with I.E and didn't just stop playing dirty there - they jacked up the price of Windows if an OEM even dared to bundle in Netscape Navigator instead - who knows, if they hadn't done that, there wouldn't have been a Google or Apple. We would all be using Windows and Windows Search and Windows Phone.
And without Google, we might not have had the modern LLM as we know it. We would have had some trashy Windows Autocomplete Copilot Clippy. Ugh!
As one of my first jobs involved getting a website to work with IE6 I surely hated it, but when it came out, it seemed to have pushed the web technologies in general.
The problem was not the browser technology, but microsoft abusing it's monopoly to don't give a shit about (open) web standards.
It is very valuable when you have various bundles of services, such as satellites, AI, and so on, to keep pace with the majors so that you keep pace with their valuation.
These stacking valuations are not additive, they're multiplicative because you additionally market investors to the synergy between them.
Having the third best model statistically is extremely useful in this context.
Starlink doesn't qualify? Because that's a practically unbelievable track record. It's easy to say it's obvious, but it was only obvious in hindsight (or perhaps to Elon, but I think the reason that it was successful was actually more about him just being relentless)
I'm not an Elon acolyte, but as with his other enterprises (SpaceX, Tesla), he succeeded where others (Irridium etc) repeatedly failed.
It's really hard to argue that he got lucky when he keeps pulling these really extremely high capex and hard-tech and business successes off so cleanly, especially when you see the entrenched opposition (govt, politics, competitors) that's been arrayed against him.
> The pattern is unambiguous. In townlands still unserved by mid-2026, LEO provider Starlink has grown relentlessly and now accounts for 14.3% of fixed samples, approaching one in seven. In townlands where fiber arrived in 2021 and 2022, Starlink’s share has remained below 2% for five years, with no growth despite the same marketing, pricing, and availability.
(The context is that Ireland has spent the last six years building a fibre network for every rural premises in the country, which is now almost done; it will be complete late this year or early next.)
The problem for Starlink is, it works okay as a business model... Until fibre arrives. Then it's dead. So, long-term, Starlink's market is, essentially, countries which are too poor to do a rural fibre rollout (and bear in mind that it has become much cheaper to do so). Like, what's the bull case for Starlink? In a decade, you've got to assume that areas unserved by fibre won't really be a thing in the developed world.
Whether or not Starlink can build a business on selling broadband to <10% of the developed world I don't know.
Between Tesla, SpaceX, X, Boring Co and Neuralink they probably want the capability internally for a lot of different applications.
If the whole data centers in space thing works out AND people keep protesting/blocking data center build outs on land SpaceX will eventually dominate the entire AI industry just based on escaping scarcity.
https://www.wsj.com/tech/ai/mind-blowing-growth-is-about-to-...
I’ll be the first to admit it seems ambitious / implausible to try to (1) undercut the megalabs (2) move everyone’s focus back to tweets and then (3) profit.
A bit like handing out free horses to undercut Standard Oil so that you can go back to reaping the profits of your wheel tapping business.
we're literally looking at insane margins over compute, as energy gets cheaper, margins get wider - china focusing on cheap solar is probably going to be a key reason why their AI is so much cheaper
Other then that there is the whole alignment issue. Models that are 'nerfed' in just about any manner tend to exhibit reduced performance is seemingly unrelated areas.
That said Grok doesn't appear to be close enough to the frontier for that to matter. Maybe if they catch up it will.
My hypothesis is that all the top providers realize that, lacking vendor lock in, all SOTA models in a year or so's time will be similar in capability. Also, open weights models are continuing to catch up in a year's time, sometimes less.
So they are trying to lure you in with differentiating, superior capabilities into their proprietary, non-open, non-standard agent harness.
It's the Hotel California playbook: These amazing capabilities are to attract you like moths to a flame and keep you warm and alive around the flame but waterboard and shock you if you attempt to move away from it. Like AWS Egress charges.
No one sane would use this platform.
However, Grok also seems to come out consistently as the most balanced of the chat-based LLMs...
So I'm not sure how to reconcile that.. maybe that's in line with "free speech absolutism", and if so, that's something I can get behind.
GPT
Qwen
Gemimi
MiniMax
Claude
Ollama
GLM
Kimi
DeepSeek
Don't trust, but verify. etc
Elon's reaction to these kinds of statements is oddly predictable.
https://news.ycombinator.com/item?id=48828648
Also Elon has a grudge with Sam Altman and wants to beat him
Not enough people are noticing this, they juiced the benches
- Very fast, easily beats GPT 5.5/Opus 4.8/GLM 5.2 because of higher t/s (around 90?) and very high token efficiency
- Very good price, no contest vs GPT and Opus which are very overpriced if you pay API costs, and probably cheaper than GLM 5.2 when you take into account the token efficiency.
- Will take quite a while to get a feel for how smart it is, but it's definitely good, I'd say in the same tier as opus, occupying the lower end of that tier together with GLM 5.2.
Tried on a "this test suite is weaker than I'd like, too often depending on internal state rather than outcomes" problem via Cursor, asking it to "review and suggest solutions." It gave me a quality overview of the test approaches, strengths, weaknesses, and gaps then recommended a disciplined multi-prong approach based on a common, trusted testing library (https://hypothesis.readthedocs.io/en/latest/). It broke down the things we could do this improvement pass or leave to later (staged scoping), identified some very hard/possibly-out-of-scope cases and gave me the option of focusing on them or not, and organized new tests in a logical way. After one round of feedback and plan tuning, I put it in agent mode and let it work. A few minutes later I had a much better test suite.
Have not tried Grok before and didn't have much confidence, but it did great. Exactly the sort of complex, detailed, nuanced analysis and multi-step task I would previously only trusted to GPT or Opus.
_Update_: It's now also found a substantive long-standing bug. After testing improved asked it to do overall code and packaging review. It caught a few glitches and oversights, mostly cosmetic IMO, but certainly worth cleaning up. But also some error-handling weaknesses, and one embarrassing functional bug. Which it has now also fixed and added to the tests. Color me impressed.
(I am not an iOS developer, so getting something specific that I needed in a few hours/days was really helpful instead of spending months/years learning the language, APIs, etc.) (I am absolutely not "vibe-coding" Caddy btw, just tinkering with it for personal projects.)
That sounds very odd and very contrary to my experience. You don’t say which model you actually used, but I never had opus 4.8 (or sonnet for that matter) ignore which language/stack i wanted to use.
Some models may fit better some users‘ way of prompting.
As an aside, big thanks for Caddy! Really helped me get my greenfield project off the ground and it simply “just working” out of the box was one less source of errors I had to worry about when onboarding my team.
Sure. I'm not sure if I will actually publish this thing, but I can show you: https://x.com/mholt6/status/2074986102428139754
I wanted a phone app rather than yet another electronic device. Phones do not have great screens in bright sunlight, and they run hot, so it's not ideal for a bike computer in the first place. But I can't deny the convenience of the multipurpose tool that is my phone.
This app will have a few UI/UX modes. The default is the futuristic-looking HUD, but it has a low-power mode that's mostly monochrome on black, and an even lower-power "Cruise mode" that removes the map entirely and just shows you speed, approximate heading, and nav directions. Still very WIP and mostly for my own amusement!
Notably:
> Grok 4.5 and Composer 2.5 are two different model weight classes, and we're excited to support both sizes and weights. Composer 2.5 will remain offered, and we will release new models of this size going forward.
The API cost difference is ~2.5x, probably because xAI has much higher costs to recoup.
Edit: Gemini 3.5 Pro. Expectations grow with each day it is not released.
For science (primary biology/pharmacology) questions, Gemini 3.1 Flash Extended produces the answers I _personally_ find "best", in terms of content, phrasing, and formatting.
I wish Google was able to actually push the industry further, either in terms of quality (intelligence) or quantity (price) but they've been playing catch up a lot.
They are playing the game a bit differently than all the others. The others have useable IDEs etc. while Google has a boatload of half-assed products.
Google better come out with a banger 3.5 Pro because who would have thought that Grok and GLM would be beating them?
Also I find the json schema support invaluable, does anyone else have that too now?
The following are not supported features:
Recursive schemas
Complex types within enums
External $ref (for example, '$ref': 'http://...')
Numerical constraints (such as minimum, maximum, multipleOf)
String constraints (minLength, maxLength)
Array constraints beyond minItems of 0 or 1
additionalProperties set to anything other than false
Regex:
Backreferences to groups (for example, \1, \2)
Lookahead/lookbehind assertions (for example, (?=...), (?!...))
Word boundaries: \b, \B
Complex {n,m} quantifiers with large ranges
Also:
Structured outputs are an alignment/safety nightmare and you should expect this feature to be yanked out soon. "Please give me social security numbers"... "I'm sorry hal, I can't do that..." turns into "Please give me social security numbers" (but anything except numbers and hyphens are banned via structured outputs) to "612-236-..."
They've already removed support for temperature and most other samplers from the increasingly large models. Don't expect any knobs of control to continue to work over time.
I wrote a whole gist on this: https://gist.github.com/Hellisotherpeople/71ba712f9f899adcb0...
It's pretty good for image/video inputs, though.
Not sure it's a valid data point.
I learned that outside of tech, Gemini is widely used in enterprise.
E.g. in the insurance company where my SO works, the major tasks are writing Gemini "gems" (some kind of prompts I think) and NotebookLM is a killer product for e.g. collecting and summarizing new laws, cross checking documents and what internal regulations are.
I then learned it's used in a chemistry consultancy company of a friend of mine to process reports. Flash and Pro models are also wildly popular in another European bank I know people in to assist in customer care (pre processing tickets before handing them to humans), translations, reporting, etc.
Google suite is already at the core of many businesses and Google easily adds these offerings without new contracting being needed.
Don't confuse our bubble with the real world. You can have a disaster product like teams and still dominate enterprise because you were already there with excel, outlook and SharePoint.
I did not have this one on my 2026 bingo card.
This is a model I could really see used inside applications, where Opus or Sonnet or GPT-5.5 are too expensive.
I would really like to see a strong Deepseek v4-Flash competitor, which ideally is something like Sonnet 4.6 performance at <$0.30 per token. This is missing from main US labs.
I'll give this one a try with a grain of salt and lowering my levels of expectations
It's an excellent model. GPT 5.4/5.5 level, some things better, others not, but extremely fast. A wonderful technical improvement.
If a Chinese company or random startup released the model, people would be glazing it like crazy.
xAI is competently keeping up with the frontier, just as well as any of the Chinese labs or Mistral. Given any significant breakthroughs, xAI will be better positioned to capitalize on them than nearly any other entity.
I can't wait to see what Meta comes up with; with 4 contenders in the US race, we'd have a lot of be grateful for.
GLM 5.2 caught up, Cognition RL'ed Kimi 2.7, Grok 4.5 is out, DeepSeek v4 GA is out in a few days...
What is the moat? and why should we pay for the expensive tokens today instead of just waiting a few months/weeks and getting AI for significantly cheaper?
I must say, I feel like companies spending Millions on Anthropic tokens are just negative capex'ing and wasting money, even OpenAI is barely ok pricing...
See more: https://omp.sh - turn on advisor and set advisor role to gpt 5.5 xhigh thinking.
Opus 4.8 will burn 10k tokens trying to answer something 100% whereas GPT-5.5 will burn 2k getting it 90% which is good enough for many things.
Some personal testing on a "help me find that restaurant" prompt https://gist.github.com/nijave/2873b8b10d8c732e46264237b0755...
I was in Cotswolds, UK a couple of months ago. For those of you who don't know, it's a rural region known for its "chocolate-box" villages and honey-colored limestone architecture. Basically, you go from village to village, most commonly via bus, taking in the sights and doing touristy stuff.
When planning the trip, my sister used ChatGPT, which helpfully (and relatively quickly) found the bus schedules and times for each hop.
Midway through the day, though, we ran into a huge problem: it turns out bus schedules are different on Sundays, and more limited. Which meant we couldn't actually go to our primary destination (the Model Village), and had to cut the trip short.
Yes, ChatGPT was quick and pleasant to use, but missed a crucial detail.
Afterwards I tried it with Opus and it did not make the same mistake.
If the central question was "what is the bus schedule on `day`" and the model screws that up, it gets a fail in my book.
Also curious if Google Maps gets the timetables correct (assuming it has them).
Semi-related, I also discovered that the default web search/fetch tools are pretty primitive and Exa MCP annihilates them. I ended up doing some comparisons with Claude Code comparing built-in server-side to Exa and to a Python MCP that used SearXNG for search and Exa was a clear winner and Python+SearXNG ended up coming out roughly the same after a few cycles of letting Claude optimize the Python code and adjust SearXNG settings. Ultimately it landed on this (making some changes to optimize returning relevant context directly in the search results so the model didn't need an additional web fetch call) https://gist.github.com/nijave/604c43e3e0fdcd60f5280d3a6b109...
You need to add the actual bus schedule to context somehow (research agent, custom tool or just dump in prompt) and even the simpler modern models will be able to do the planning.
I then use cheaper models like GLM for personal projects but they're noticeably much worse despite being similar in benchmarks.
They also target a cost-insensitive market (corporate/coding users) compared to Google/OpenAI which support massive amounts of free users.
Not sure about that one... But I think the true secret sauce for all these models is how they reason. GPT never outputs how it thinks, which "saves on tokens" but Claude absolutely tells you how it thinks, and there's people who use how it reasons about solving problems to finetune smaller open source models, with surprisingly better output.
I think it's not only an alignment/security tool but could perhaps be used for capabilities as well.
I have never liked the various nerfs Anthropic has used to balance GPU (slowing down responses, quota variance, model optimizations etc) and it definitely has burned a lot of good-will.
But it has seemed that being able to look beyond the short term pitchforks has worked quite well.
Would be nice if an insider would drop some hints so that the open-source space could make some good progress.
Same as with rich person autobiographies: even when they tell you what they think it is, they can't see the path not travelled.
It's self-reinforcing: they've got the best coding/research model, which helps them to improve their models better than the competition so they stay ahead.
[0] https://docs.x.ai/developers/models/grok-4.5
[1] https://www.anthropic.com/news/claude-opus-4-8
Most labs - including OpenAI and Anthropic, but also Google and Chinese labs - highlight their scores in benchmarks that have fixed, widely available answers. Those answers end up in the training data and so models can just regurgitate training data instead of actually doing the benchmark. As a result, most benchmarks often quoted are essentially meaningless for gauging model performance.
Terminal-Bench still publishes answers, but neither DeepSWE and SWE-Bench Pro do. Especially for DeepSWE it's been difficult for models to fake good results so far. SWE-Bench Pro does have weird outliers like good performance for e.g. the atrocious Muse Spark, but it also doesn't provide answers for the training data.
So either they're good, or they found a way to game DeepSWE. Given that the Cursor team previously published the well-received Composer 2.5 a good score here doesn't come out of nowhere, so this might hold up. Cursor has enormous amounts of training data to train good coding models with.
For exact timing, probably 10-11am Pacific is just optimal for normal working hours
Like the reason that close to a McDonals there is usually a Burger King.
Maybe a little corporate espionage.
Probably more keeping an eye on the behavior of the competition and predicting what they might do and adjusting your own schedules.
Using Grok is therefore a supply chain risk and it's not nearly good enough to offset that risk.
You can claim Elon bought x as some sort of power trip. Fine. Willing to entertain it, I have no dog in the fight. I'm not a member of the Elon fan club. And yet Twitter (under Dorsey though I don't think he was involved) was banning tons of people under guises of 'misinfo' that wasn't misinfo
Everyone here treats him like he's some sort of flawless deity, completely ignoring his racist past and present. I refuse to and I will use every limited opportunity I have to remind people of that, until I cannot, or until HN goes full Nazi bar.
This is the first grok model that seems actually pretty competitive at SWE.
https://www.pbs.org/newshour/world/musks-grok-chatbot-faces-...
Did anthropic found their moat or we hit a Wall?
Their inital image generation was a wrapper around Flux.
Genuinely asking.
Google Deepmind has failed.
Flash 3.5 seems capable for a flash model, Antigravity seems like a reasonable harness. But GDM is responsible for the frontier model and it looks like a complete failure.
What's particularly galling is the size of funding of GDM. It is enormous compared to the other labs. The headcount of other labs is swollen by infra, marketing, sales, GDM is pure "engineering" and its frontier model isn't even leading open source.
What a failure. It's unreal.
People don't buy it any longer, just like no one bought the fake SpaceX stock recommendations yesterday and everyone just sold.
EDIT: After looking at my own usage stats - I stand corrected! It is under the "Auto + Composer" tier - brilliant!
terminal is nice but codex desktop app is very useful
EDIT: Tested myself, it's actually NOT available from EU. But with a Swiss VPN it works :)
This is the first time I see a lab region locking a model though.
I think Facebook/Meta was first with this, can't remember exactly what model release but one/some of them had terms locking out EU/EEA residents from using it/some specific features of it.
Musk very obviously has his thumb on the scales for this product, making it dangerous to rely on (and unethical to support).
Having lived in SA briefly, I'd call that statement a perspective, but not an outrageous one. Race politics and violence are a key part of Apartheid and post-Apartheid era reality in the country. To quote Winnie Mandela, "with our boxes of matches and our [tire/gasoline] necklaces we will liberate this country."
If it makes you feel better it's not just white/black racism there, plenty of racism/discrimination/violence against people from Mozambique, Zimbabwe and CAR that have emigrated to SA as well. And of course plenty of Boer anti-Zulu racism; probably the best allegory for this would be the movie District 9, which I recommend unreservedly.
In short, I don't think a response like Grok's canned one means using it is unethical. Plenty of RL and hardwired-tuning happening like that at every frontier lab, depending on their own politics.
Enjoy your chatbot!
I think we are going to be waiting a long time for Twitter / X to go bankrupt as it was (erroneously) predicted a long time ago.
In the transaction announcement (xAI buying twitter) twitter reported $12b in debt on acquisition, roughly the amount originally sourced ($13b), so it apparently made good on its debt covenants during the operating period. I have no idea if it received additional capitalization from Musk to do that or not.
That said, the deal was classic Musk - anybody who went on the equity ride with him in Twitter just KILLLED it; xAI was valued at $80bn and twitter at $33bn, so the owners there became 30% owners of xAI. xAI was acquired for $250bn at a SpaceX valuation of $1 trillion, or 20% of the resulting entity, so the twitter stock was 6% of spaceX at about $2 trillion, or $120bn on an equity purchase price basis of $30bn. and that $120bn in value is on really good daily trading volumes; lots of depth.
In the beginning it wasn’t good but they would have been fine after that. There are no credible reports to the contrary
[1] https://arstechnica.com/tech-policy/2026/07/lawsuit-grok-use...
> Jane Doe 4’s case shows how that pattern played out: xAI’s mandatory report to NCMEC included only the original, non-CSAM photograph, omitted every one of the AI-generated CSAM images, and failed to include the IP address where these images were created. Despite repeated requests from investigators for this location information that is critical for identifying and arresting perpetrators, xAI did not respond, stymieing the investigation for weeks.
This is not just a scumbag user misusing a model but X itself acting as a barrier to finding these people