You may be new here. This is Simon’s de facto benchmark for models. I happen to find it a really good one.
Small aside: It’s crazy to me that while it’s improved over time it does seem like most of the models haven’t been trained specifically to defeat this one.
This supposedly is better than KimiK2.7, as much hype as GLM5.2 gets, I find myself using KimiK2.7 half of the time, so if the benchmark is true, then this can definitely go in the mix. My hope is that it might have strengths in some areas to beat all other open weight models.
North Mini Code by Cohere (HQd in Toronto) has honestly been very competitive in my personal assessment with many of the models coming out of the PRC. I'd position it below Moonshot AIs and Z.ais recent releases, but above the varieties of Qwen, Deepseek, MiMo, etc.
Depends whether America the continent or just the United States counts of course.
It's not the better model since Llama3. Trinity Large is American and quite decent, unfortunately tons of crazy good models have been out and it's harder to run locally at 400B. I think Arcee did a terrible job of promoting their model.
British-American with much of the research happening in London. I don't know if it's known what team specifically worked on which of the Gemmas, I'd suspect it's a healthy mix of multiple satellites across the globe, but the contribution by the UK Deepmind parts likely was significant, certainly not so small that it'd justify downing a regular question.
You don't hear about them much because their models aren't really competitive. I really wanted to try Trinity Large as a daily-driver in the MiniMax M2 sort of niche but I couldn't make it through a single day. The models need another couple point releases worth of post-training to make useful agents and if memory serves they weren't any less slopped in writing style and those are really the only two things people look for in models.
It could be but there are a host of companies going after open weights models: Arcee, Reflection, Llama (TBD on Meta's focus on closed-source versus open-source), etc.
That said, the fine-tuning API + open weight model at least is a semblance of a viable business that could work so I will be curious about it. I'm not sure the synergy is fully there (why is someone with an open weights model privelaged to fine-tune it better if it's just QLora or Lora) but let's see!
Thinky's main commercial product AFAIK is Tinker [0] - companies pay them to host their fine-tuning workloads and then the resulting fine-tuned models. I don't know if this is a good business plan, but I'm sure at least one person there has read Joel on Software [1].
> It could be but there are a host of companies going after open weights models: Arcee, Reflection, Llama (TBD on Meta's focus on closed-source versus open-source), etc.
my bet is that Chinese government fund Chinese models way more compared to what those companies receive (except llama, which is outdated but was strong foundation at its time)
The story of Reflection AI is supposedly that the company was faffing and failing at winning in the coding agent space, but was introduced to Jenson, who suggested they build an open-weight model and said he would fund it. That turned into a $2 billion financing with NVIDIA doing roughly $500 million and was a complete pivot.
I think the bet would have to be that a US Open Weight company either: 1. Gets a lot of money from Jenson who views them as a counterbalance to the big labs in his ecosystem and a way to generate leverage (the same way he is positioning neoclouds-- it also could be synergistic with neoclouds who could offer the model serving endpoints) 2. Can fast follow the same way Mistral does (which, honestly, seems like just distilling the Chinese model, which distills the US lab but is pretty innovative on a whole lot of architecture both in training and serving land.) 3. AND figure out some (maybe not super lucrative but lucrative enough) sort of business model, as well. There are lots of possible business models, so I will be curious how this whole space evolves.
>The story of Reflection AI is supposedly that the company was faffing and failing at winning in the coding agent space, but was introduced to Jenson, who suggested they build an open-weight model and said he would fund it. That turned into a $2 billion financing with NVIDIA doing roughly $500 million and was a complete pivot.
"Current GPT/Claude/Gemini" is not a meaningful statement about perf. There's many different models from each of those providers and there's a considerable gap between the best of anthropic and open ai compared to gemini.
Benchmarks have GLM 5.2 somewhere underneath Sol and Fable and closer to now last-gen openai and anthropic models.
AllenAI is also one to keep your eye on. Founded by Paul Allen of Microsoft, they are one of the best teams working towards truly transparent / open AI (including training data)
Open-source models + services. This is more attractive because it doesn't lock in the vendors. If I grow larger, I can decide to deploy the open-source models.
Better than average chance I’d say. I suspect they are hoovering up EVERYTHING that gets sent to them. Whether that’s a problem or not depends on your data. I do wonder how many security tokens they get in the stream on a daily basis.
So they're constantly hemorrhaging their most valuable clients?
Tech history is littered with the corpses of "open source but we sell hosting" services. Models are so expensive to train, you can't be losing the big clients once they get super profitable.
This is genuine, noob question: how is this different from AWS?
I get that they're in very different businesses, but for both don't they have the issue that once a client gets big enough the client might decide to move the services in-house? Based on how much of the internet went down when that AWS data center crashed the answer is clearly "No" for AWS.
Is that because of physical, real-world infrastructure? Are there no open versions of their APIs? Is it too hard to migrate to something else once a client has achieved that size?
I would say "it's risky and requires a lot of labor to migrate without corruption, loss of data" and also minimizing downtime. Sure anyone can run pg_backup, but can you do it across 90 databases? Can you do it live? Can you coordinate rollout of the process, cutover, and monitor for failure? What's the cost of egress for this? Is the team your A-team or the B-team? Can you trust this to the B-team? Is it worth having this team spend all this time on a migration rather than, say, getting something new set up, or optimizing performance on an existing system?
I'm a database guy, but the same migration argument is presumably also extra work for (say) blob storage, networking, etc.
Since LLMs are stateless by their current implementation, switching to "the same open-weight model running in a different datacenter run by a different vendor" is "just" switching the API endpoint. (If they are the exact same shape, it's fine, if they differ somehow, there's perhaps some work to do there, fixing things and monitoring for failures on switch-over)
There are several open APIs it seems and OpenRouter.ai is doing a fine job making a commodity out of models and datacenters.
I don't think it's that difficult. Their servers are stateless too. S3 is easy to migrate.
Database is more difficult, but tons of people have done it successfully.... meanwhile people who host their own LLMs are relatively small in number in comparison.
Most companies don't do their own data centers mainly because it is more expensive and less reliable. It's something they can just pay for the problem to go away. The calculus for hosting your own LLM is probably similar.
Even Stripe who built their own coding agents and has tons of money/resources still decides not to host their own LLMs.
Still, many people will prefer open-weight models. It is similar to how we prefer linux but still use AWS/Render/and whatever. It doesn't lock us in, and we can move providers if we want to.
To compete against America. If your country has something like DeepSeek you really can't afford to let it fall as it's your best leverage if the US government decides to ban companies in your country from accessing American LLMs. And this is why there will never be a "DeepSeek of the US."
Considering how volatile things can get depending on who's president, I'd say even American companies need to "compete against America" if they don't want to get their rug pulled from under them (which, apparently, the legal system allows to easily happen in the US).
Thinky has a potential answer in Tinker — give away the weights and charge for the SFT (and maybe RL down the line) to make the model more capable for specific tasks.
It doesn't matter until it does. If the chinese government decides that open weight model releases are no longer allowed, that's a lot of companies that can't release new models. Same with the US government, etc. Having diversity is important.
It's a similar problem the human DNA solved by telling our teenage selves that our parents are dumb and we needed to move to a new tribe. Genetic diversity, but a digital equivalent.
China’s got absolute control over its outputs. For America to have any guarantees around long-term availability of OW models, they need domestic production.
FWIW this is the same logic for China’s need for their own OW models
I think practically every government will want to put restrictions on private companies building models.
Frankly the EU and the US will practically be less involved and have more pushback from the public in this than China. I think that’s less “China bad” than recognizing that China is a more authoritarian state and has far more proclivity to interfere than western states.
Maybe I’m wrong? What does deep seek say about Tiananmen square in 1989?
Its not as good as GLM 5.2 for agentic workflows while also being bigger. Competition is going to be ruthless because the super low cost to switching.
There is also AllenAi in the US, but they have yet to produce a model at this scale. Thankfully, new contenders can come out of nowhere and do well, as long as they can produce a competitive model.
> Its not as good as GLM 5.2 for agentic workflows while also being bigger
GLM 5.2 underwent extensive post-training and iteration since its original release to reach its current state. This seems like an extremely strong model for a first release, with a lot of potential for improvement, just like DS4.
Sometimes I wish Meta had stuck with Llama 4 a bit longer to see how much further it could be pushed.
Llama 4 wasn't deemed a success, and Meta pivoted away as its now former head of AI couldn't demonstrate, nor even showed interest in, business profit.
They overspent on llama 3 anyway so money ran dry, LeCun is good at running research, but budgets didn't stretch. Meta isn't investing in frontier big models anymore.
> Inkling is not the strongest overall model available today, open or closed. Instead, a combination of qualities makes it a good open-weights base for customization: multimodal capabilities, efficient thinking, and availability on Tinker for fine-tuning.
Open base models that can be fine tuned on Tinker is a great business model IMO. You (i.e. an enterprise) can own your own model & have it perform frontier-or-better at your task at potentially much lower cost and Thinking Machines gets to be your essential infra/service provider in this world.
Also,
> Inkling-Small matches or exceeds its larger sibling on many benchmarks — the result of improvements we made to the pre-training data and recipe for the smaller model.
Very cool! Excited to see the next generations of Thinky models.
What strikes me the most is just how many different tasks are involved in modern model design. It used to be the case that you come up with a new loss function, slight architecture changes, etc., run your train and eval loop, and publish the artifacts.
Now, there’s so much work to do just to keep up. It’s the ultimate red queen race. All of the 500 steps involved, each of which is its own little optimization loop, is sort of awe inspiring.
But obviously this inverts the previous rules that small teams run faster than big teams. AI requires a big team. It’s only once the team pushes past the 1000s that organizational inertia seems to become an issue. Because until then, there’s way too many pieces for even a dozen super stars.
It's nice to see a strong long context open weights model that is multi-modal.
There are many applications that will benefit from the strength in audio here and until z.ai and co work in visual this could be very strong for general agentic applications, though I see there's a bit of weakness in the benches for areas that might make that less true.
Like all models need to slap it in your harness and do proper evals on the tasks you care about.
MiniMax M3 and DeepSeek v4-Pro are highly capable long context open weight multi-modal models. But long-context is a trap, because performance still falls dramatically after 150k-200k context.
> But long-context is a trap, because performance still falls dramatically after 150k-200k context.
I often see this repeated, and it is not true task to task. I work on this daily and we have several tasks where long context is advantageous and our evals against a whole battery of models with different windows show it as being so.
This is why having good evals for the tasks you're working on is so important.
> But long-context is a trap, because performance still falls dramatically after 150k-200k context.
I'm not sure exactly what causes the difference, but this heavily depends on the model. In my experience with Opus 4.8, I can go well over 500k and still get extremely good results. A drastically different example was GLM-5.1, which worked great until about 100k and then turned insane almost immediately. They did fix that with 5.2, though.
Very preliminary testing so far, but there is something here, far beyond what the benchmarks suggest. Only ever saw such outperformance of public evals vs my private ones with Anthropic models and while it is far to early to make any judgement at this stage, this model will take up a lot of mine time in the coming weeks by the look of things. Only ever viewed Moonshot AIs models as something I'd be able to live with open-weight-wise (Z.AIs output simply does not perform as well in my task set), but this has the potential to be the second. If Mistral came out with something like this, I suspect every Europhile (me included) would never stop talking about it.
Quick and still very early update, the model has (with web search disabled which was verified via the reasoning traces) accurately answered a number of questions focused on very niche details (engine specific maintenance in certain newtimers, very niche bag construction and material details) that I have only ever seen Gemini 3 and 3.1 Pro get correct. Neither Fable 5, nor GPT-5.6 Sol or any other model by any other lab has ever provided accurate information without web access for these specific questions for which an objectively correct answer absolutely exists and is general knowledge if one is versed in the specifics.
Being ahead of Fable 5 in any task, that is not included in public benchmarks and thus could be overfitted for, is impressive to say the least. Last time a model exceeded the expectations I had based on the release notes to such an extent was Haiku 4.5, which I still wish we got a solid replacement for.
For a first model, and given it's open, I am gaining some faith in American Open research labs again...
I couldn't test it since it's not on openrouter or something, but even if it's only as good as GLM5.1 that's more than good enough first attempt, I think.
Perhaps a lot more labs will catch up to ballpark frontier esque level soon, I am all for more competition in any field.
Seems like this is particularly good at instruction following, but not as strong at coding as others. It's always great to get more diversity of open weight models though! I'll need to test this out to see what its "personality" is like.
This is the best voice/tone I've seen from any model so far. It's using filler words and phrases in places that normal people would put them, rather than sounding like a corporate customer support agent!
It's nearly double the size of Nemotron 3 Ultra, so I'd expect it to be considerably better, although the active parameter count seems to be a touch lower at 41B vs 55B
For thinking machines, they provide super simple finetuning APIs.
if it is their model, they can have more lower level integrations for that.
Thinking machines might be the only large lab in the US to have business interest aligned with open sourcing strong models that are customizable.
Just serving the model over API seems like a natural fit and is what many of them are doing. So simply being the cloud provider for your own open weight model can be a source of revenue
But so can everyone else. What’s the moat for spending all those billions. I understand the Chinese angle, they need to undermine American models as a matter of statecraft, but what is the business model here? It just seems like VC charity.
What is the moat? The time it takes for AI to rewrite an efficient inference stack for a new model? Considering most LLMs follow a similar architecture, adapting to a new model shouldn't take that much time.
There is no moat. At the moment, all of these companies are burning money to gain mindshare and market share. That's what Thinking Machines is doing; they're not looking for a business model.
I don't know why people keep saying there's no moat. There's no moat. Having a FUCK ton of money to train these gigantic fucking models and retain the brains to make it happen is a moat.
You're not going to train one using a VPS from LowEndBox.
Open source low cost models will dominate most enterprise tasks as cost curves will dictate usage. TM is trying to replicate that especially as the US and China gets more defensive with their tech
Similar to companies working on FOSS codebases, hosting (sometimes with the license restricting third-parties in some way), providing tailored models and services to customer's and getting bought for your team if your model happens to be competitive enough.
I never thought i'd see the day they released a model, rather than a blog post. The Figure 3 demo being a screencap of chrome in localhost made me feel better about myself. Jokes aside, best western open weights model- very cool.
They are one of the few labs (perhaps even the only one at this level) that are doing something both unique and useful, rather than simply imitating what the others are doing: https://thinkingmachines.ai/blog/interaction-models/
Interestingly, when opening this page, the first thought I had was not that the benchmarks should be high, but 'I really hope they did not benchmaxx'. I think a model with modest benchmark scores can have much better real world utility as opposed to the current frontiers that are RL'd into being robotic and rigid.
Excited to try out its capability, especially audio and video.
It's nice that it has a long context window, but in practice, I find I always have to clear context btw 150k-400k context even if the context window is 1M on paper.
They also indicate they have a 276B A12B version, but it doesn't seem the weights are available. This might actually be able to fit in 128GB when quantized to 2 bits or so which makes it interesting.
They mention in the announcement link https://thinkingmachines.ai/news/introducing-inkling/ that they are still testing Inkling-Small and it will also still be multimodal. This makes it super interesting as a Deepseek V4 Flash replacement (and would be interesting with DwarfStar / ds4 if it gets supported).
Interested in the implied strategy - that training a bespoke model for what you need will make economic sense over using a mass-trained model. I wonder if that's true?
Same. Gutsy bet to make in the face of Fable / Mythos, but the multimodal quality is at least a promising technical/ product story to tell. Everyone knows throwing Opus at everything is wasteful and domain expertise should live in the weights eventually; the question is whether foundation model scaling will slow down enough soon enough for that to matter.
Or maybe this is just a warm-up / stopgap and Thinking Machines is betting on finding the next architectural breakthrough that lets it compete with the big foundation models?
the open-weight model release cadence is approaching npm package velocity. soon we'll have left-pad-7b and someone will unpublish it and break half of production
I think we’re going to start seeing more OSS models that perform especially well on certain tasks instead of trying to be generalists like the frontier models. That’s a winning formula because if you’re building an app on a model it often has a specific set of use cases
We always have been, Big Tech has been extremely slow to catch up to the indies.
Nobody is making money lmao.
I would not bet on OpenAI creating any good products, they never have. They are like Meta in all this - never innovated anything themselves, can only acquire others to stay relevant. They'll never do an incredible consumer experience on the level of a PlayStation or Blizzard or even Google.
competition in this space is great, especially with open models/weights. I think the answer is not closed source models. Similar to the Unix versus Linux situation in the 1990's, open source wins out. Yesterdays story about how OpenAI has now began encrypting traffic between model and agent [0], this story brings a breath of fresh air. There is nothing "Open" about hiding the communication between model and agent, especially with software that is running within a trusted environment/network. It needs to be more transparent, not less.
I really respect the epistemtics work here. It might become an accurate, inexpensive open-weight workhorse for high-level prioritization and decision-making work. (Finance bros will also love this)
> If it's ~30% bigger and not as good as GLM 5.2, why would I tinker with this model?
The benchmarks never tell the full story. Some of the open weights models have been benchmaxxed for a while. Their utility on real work can be different than the benchmark number.
The multimodal input is also a big deal. Having vision input is really helpful for a lot of tasks.
I second that. Gemini 3.5 Flash rocks the benchmark charts but is terrible as an agent. Horrible instruction adherence and makes WAY too many tool calls
I'm not sure why I'm being downvoted but I didn't mean it in a negative way.
For such announcement, I would expect them to give me clues on when I should use this model and in which cases it's the best one.
The benchmarks that they share doesn't indicate that it's cheaper to run than other models, or can fit in my local machine, or excels in a specific vertical.
After reading the comments here and X, I can see it being the top-3 multi-modal open-source model though.
If they have a really seamless fine-tuning experience and maybe can help you extract the data you need to FT (which is one of the big challenges in actually getting fine-tuning democratized), maybe you would use it because "Tinker" defaults to it.
The model could also be more flexible for non-coding use-cases (they show the results for reasoning being strong) so maybe the argument is to use it for non-coding use-cases to drive relatively deterministic conclusions for non-coding agents (they have also done some determinism work on kernels, which could be useful in pulling on that thread of deterministic models that are fine-tuned for everything that is not writing code.)
That said, I'm not sure how much all the work they have done actually synergizes or if the market size (at least in the short to medium term) is big enough for a huge outcome from the company's current valuation with those bets as the enterprise agent estate is taking a while to evolve. Hence companies like Anthropic and OpenAI are throwing tons of consulting money at the problem.
There's also an Inkling-Small that is 276B, 12B active that is much smaller than GLM 5.2 and still multimodal. Not released yet, but in the announcement link they mention that they're testing Inkling-Small & will release as open weight after testing. That one may be interesting as a Deepseek V4 Flash replacement.
Is it really that bad? I always get the impression that their blog posts look especially beautiful with their font choices and overall design. They are typographically pleasing, and if I could, I would use this as the distraction-free reading mode for every web page.
It feels like I’m reading a newspaper, but oddly, without them resorting to any skeuomorphic tricks.
Raised 2 billion dollars at a 12 billion valuation and debuts at 41 on the Artificial Analysis Intelligence Index, while KIMI and DeepSeek will release Fable-class models this week. What a joke.
Moonshot (Kimi) has raised $3.77B and been around for >3 years, Thinking Machines raising $2B and releasing a decent open weights model in 16 months is actually quite comparable.
> ...while KIMI and DeepSeek will release Fable-class models this week.
What new model is DeepSeek releasing? Their current V4 Pro at Max reasoning is consistently worse than GLM 5.2 at Max reasoning, though the latter is close to Opus 4.8 at Extra/Max reasoning, albeit a little bit worse in my experience (though if they gave comparable amounts of tokens to Anthropic 5x Max subscription I could see myself moving over, currently they give you less though even with their ZCode discount).
In practical agentic development, none of those seem to be that close to Fable to me. Spent 181 million tokens with GLM 5.2 with ZCode in the past month, 142 million with DeepSeek V4 Pro with ZCode and OpenCode and about 3.45 billion across all Anthropic models with Claude Code, though understandably with my workload between 95-99% of them are cached (very docs/plan/tooling/read heavy work to limit slop, albeit with sub-agents and workflows).
DeepSeekV4 was a preview model, read the papers. It's not the final model. They released it to demonstrate architectural capabilities. They are still training and the model release is planned within the next month.
I haven't used Fable, but if the hype is to be believed then it's a jump in model capability. If so then I don't expect the next DeepSeekV4 version to match it. However, if the next DSV4 version get's the kind of jump 3.1 got over 3.0 or 4 got over 3.2, I'll be very happy with it. Progress is progress. We "can" run DSV4 locally, Fable is closed.
Small aside: It’s crazy to me that while it’s improved over time it does seem like most of the models haven’t been trained specifically to defeat this one.
If you want to run locally, checkout https://github.com/danielhanchen/llama.cpp/tree/add-inkling https://unsloth.ai/docs/models/inkling https://huggingface.co/unsloth/inkling-GGUF https://huggingface.co/unsloth/inkling-NVFP4
This supposedly is better than KimiK2.7, as much hype as GLM5.2 gets, I find myself using KimiK2.7 half of the time, so if the benchmark is true, then this can definitely go in the mix. My hope is that it might have strengths in some areas to beat all other open weight models.
Depends whether America the continent or just the United States counts of course.
https://www.arcee.ai/blog/trinity-large
Thinking Machines might be it.
Here are some of their current open weight offerings: https://www.arcee.ai/open-source-catalog
That said, the fine-tuning API + open weight model at least is a semblance of a viable business that could work so I will be curious about it. I'm not sure the synergy is fully there (why is someone with an open weights model privelaged to fine-tune it better if it's just QLora or Lora) but let's see!
[0] https://thinkingmachines.ai/tinker/
[1] https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/
1. Magic
2. Managed hosting of their model
3. Hurting competitors. If people are using Meta’s commoditized models they’re not paying Google or allowing OpenAI to become too big.
4. Free R&D from open source. If open source developers are optimizing systems to run Llama, that helps Meta.
5. More magic
The Chinese "Neijuan" aside, most competing labs are going for the classic, 'your margin is my opportunity': https://tomtunguz.com/is-your-margin-my-opportunity-software... / https://archive.vn/5Vmq3
my bet is that Chinese government fund Chinese models way more compared to what those companies receive (except llama, which is outdated but was strong foundation at its time)
I think the bet would have to be that a US Open Weight company either: 1. Gets a lot of money from Jenson who views them as a counterbalance to the big labs in his ecosystem and a way to generate leverage (the same way he is positioning neoclouds-- it also could be synergistic with neoclouds who could offer the model serving endpoints) 2. Can fast follow the same way Mistral does (which, honestly, seems like just distilling the Chinese model, which distills the US lab but is pretty innovative on a whole lot of architecture both in training and serving land.) 3. AND figure out some (maybe not super lucrative but lucrative enough) sort of business model, as well. There are lots of possible business models, so I will be curious how this whole space evolves.
You can pretty much remove the supposedly here
I suspect 2B is not enough to boostrap frontier model from the scratch (for both talent and hardware)
GLM-5.2 is the best in that class right now. It is competitive with current GPT/Claude/Gemini.
Benchmarks have GLM 5.2 somewhere underneath Sol and Fable and closer to now last-gen openai and anthropic models.
I find it wonderful that, as a non-profit, they are only one to two years behind SOTA models that cost billions of dollars to build, if not more.
Open-source models + services. This is more attractive because it doesn't lock in the vendors. If I grow larger, I can decide to deploy the open-source models.
there is a chance their business model is absorbing government funding..
Tech history is littered with the corpses of "open source but we sell hosting" services. Models are so expensive to train, you can't be losing the big clients once they get super profitable.
I get that they're in very different businesses, but for both don't they have the issue that once a client gets big enough the client might decide to move the services in-house? Based on how much of the internet went down when that AWS data center crashed the answer is clearly "No" for AWS.
Is that because of physical, real-world infrastructure? Are there no open versions of their APIs? Is it too hard to migrate to something else once a client has achieved that size?
I would say "it's risky and requires a lot of labor to migrate without corruption, loss of data" and also minimizing downtime. Sure anyone can run pg_backup, but can you do it across 90 databases? Can you do it live? Can you coordinate rollout of the process, cutover, and monitor for failure? What's the cost of egress for this? Is the team your A-team or the B-team? Can you trust this to the B-team? Is it worth having this team spend all this time on a migration rather than, say, getting something new set up, or optimizing performance on an existing system?
I'm a database guy, but the same migration argument is presumably also extra work for (say) blob storage, networking, etc.
Since LLMs are stateless by their current implementation, switching to "the same open-weight model running in a different datacenter run by a different vendor" is "just" switching the API endpoint. (If they are the exact same shape, it's fine, if they differ somehow, there's perhaps some work to do there, fixing things and monitoring for failures on switch-over)
There are several open APIs it seems and OpenRouter.ai is doing a fine job making a commodity out of models and datacenters.
Database is more difficult, but tons of people have done it successfully.... meanwhile people who host their own LLMs are relatively small in number in comparison.
Most companies don't do their own data centers mainly because it is more expensive and less reliable. It's something they can just pay for the problem to go away. The calculus for hosting your own LLM is probably similar.
Even Stripe who built their own coding agents and has tons of money/resources still decides not to host their own LLMs.
Still, many people will prefer open-weight models. It is similar to how we prefer linux but still use AWS/Render/and whatever. It doesn't lock us in, and we can move providers if we want to.
If you understand the world through a Chinese LLM, you are seeing it through a biased lens stemming from biased training data.
(Also, in that way, having all major LLMs developed by the US carries a risk too. We need more diversity than just the viewpoints of the US or China.)
FWIW this is the same logic for China’s need for their own OW models
Frankly the EU and the US will practically be less involved and have more pushback from the public in this than China. I think that’s less “China bad” than recognizing that China is a more authoritarian state and has far more proclivity to interfere than western states.
Maybe I’m wrong? What does deep seek say about Tiananmen square in 1989?
This is what Deepseek replied when I asked it with a burner account. Claims it doesn't have it in its training data... sure.
There’s also Prism
There is also AllenAi in the US, but they have yet to produce a model at this scale. Thankfully, new contenders can come out of nowhere and do well, as long as they can produce a competitive model.
GLM 5.2 underwent extensive post-training and iteration since its original release to reach its current state. This seems like an extremely strong model for a first release, with a lot of potential for improvement, just like DS4.
Sometimes I wish Meta had stuck with Llama 4 a bit longer to see how much further it could be pushed.
They overspent on llama 3 anyway so money ran dry, LeCun is good at running research, but budgets didn't stretch. Meta isn't investing in frontier big models anymore.
Open base models that can be fine tuned on Tinker is a great business model IMO. You (i.e. an enterprise) can own your own model & have it perform frontier-or-better at your task at potentially much lower cost and Thinking Machines gets to be your essential infra/service provider in this world.
Also,
> Inkling-Small matches or exceeds its larger sibling on many benchmarks — the result of improvements we made to the pre-training data and recipe for the smaller model.
Very cool! Excited to see the next generations of Thinky models.
Now, there’s so much work to do just to keep up. It’s the ultimate red queen race. All of the 500 steps involved, each of which is its own little optimization loop, is sort of awe inspiring.
But obviously this inverts the previous rules that small teams run faster than big teams. AI requires a big team. It’s only once the team pushes past the 1000s that organizational inertia seems to become an issue. Because until then, there’s way too many pieces for even a dozen super stars.
There are many applications that will benefit from the strength in audio here and until z.ai and co work in visual this could be very strong for general agentic applications, though I see there's a bit of weakness in the benches for areas that might make that less true.
Like all models need to slap it in your harness and do proper evals on the tasks you care about.
I often see this repeated, and it is not true task to task. I work on this daily and we have several tasks where long context is advantageous and our evals against a whole battery of models with different windows show it as being so.
This is why having good evals for the tasks you're working on is so important.
I do grant it's a good rule of thumb.
I'm not sure exactly what causes the difference, but this heavily depends on the model. In my experience with Opus 4.8, I can go well over 500k and still get extremely good results. A drastically different example was GLM-5.1, which worked great until about 100k and then turned insane almost immediately. They did fix that with 5.2, though.
Being ahead of Fable 5 in any task, that is not included in public benchmarks and thus could be overfitted for, is impressive to say the least. Last time a model exceeded the expectations I had based on the release notes to such an extent was Haiku 4.5, which I still wish we got a solid replacement for.
I couldn't test it since it's not on openrouter or something, but even if it's only as good as GLM5.1 that's more than good enough first attempt, I think.
Perhaps a lot more labs will catch up to ballpark frontier esque level soon, I am all for more competition in any field.
> look at today's hackernews frontpage and generate me a daily briefing report (create an artifact) to read later for today's nerd news
https://chat.home.jake.town/artifacts/019f679d-99e5-7000-b02...
if it is their model, they can have more lower level integrations for that. Thinking machines might be the only large lab in the US to have business interest aligned with open sourcing strong models that are customizable.
One of the worst case scenarios regarding LLM's is monopoly control, so these billionaires know they need to invest in competition.
You're not going to train one using a VPS from LowEndBox.
Open source low cost models will dominate most enterprise tasks as cost curves will dictate usage. TM is trying to replicate that especially as the US and China gets more defensive with their tech
- RLaaS (Tinker, or the more involved FDE motion a la Reflection / Applied Compute)
Something on that level but multi-modal would be quite nice!
"But it's worse than Mistral 7b"
cape
It's nice that it has a long context window, but in practice, I find I always have to clear context btw 150k-400k context even if the context window is 1M on paper.
Self fine tuning like that though seems like a whole new set of possibilities unlocked.
https://artificialanalysis.ai/models/inkling
Or maybe this is just a warm-up / stopgap and Thinking Machines is betting on finding the next architectural breakthrough that lets it compete with the big foundation models?
Holy flashbang.
That what makes this a (potentially) safer model to build on top of
Nobody is making money lmao.
I would not bet on OpenAI creating any good products, they never have. They are like Meta in all this - never innovated anything themselves, can only acquire others to stay relevant. They'll never do an incredible consumer experience on the level of a PlayStation or Blizzard or even Google.
[0] https://www.theregister.com/ai-and-ml/2026/07/15/openai-hide...
Maybe for the multi modal?
The benchmarks never tell the full story. Some of the open weights models have been benchmaxxed for a while. Their utility on real work can be different than the benchmark number.
The multimodal input is also a big deal. Having vision input is really helpful for a lot of tasks.
For such announcement, I would expect them to give me clues on when I should use this model and in which cases it's the best one.
The benchmarks that they share doesn't indicate that it's cheaper to run than other models, or can fit in my local machine, or excels in a specific vertical.
After reading the comments here and X, I can see it being the top-3 multi-modal open-source model though.
gives me hope that the training moat is even smaller than we thought
The model could also be more flexible for non-coding use-cases (they show the results for reasoning being strong) so maybe the argument is to use it for non-coding use-cases to drive relatively deterministic conclusions for non-coding agents (they have also done some determinism work on kernels, which could be useful in pulling on that thread of deterministic models that are fine-tuned for everything that is not writing code.)
That said, I'm not sure how much all the work they have done actually synergizes or if the market size (at least in the short to medium term) is big enough for a huge outcome from the company's current valuation with those bets as the enterprise agent estate is taking a while to evolve. Hence companies like Anthropic and OpenAI are throwing tons of consulting money at the problem.
Yeah
It feels like I’m reading a newspaper, but oddly, without them resorting to any skeuomorphic tricks.
What new model is DeepSeek releasing? Their current V4 Pro at Max reasoning is consistently worse than GLM 5.2 at Max reasoning, though the latter is close to Opus 4.8 at Extra/Max reasoning, albeit a little bit worse in my experience (though if they gave comparable amounts of tokens to Anthropic 5x Max subscription I could see myself moving over, currently they give you less though even with their ZCode discount).
In practical agentic development, none of those seem to be that close to Fable to me. Spent 181 million tokens with GLM 5.2 with ZCode in the past month, 142 million with DeepSeek V4 Pro with ZCode and OpenCode and about 3.45 billion across all Anthropic models with Claude Code, though understandably with my workload between 95-99% of them are cached (very docs/plan/tooling/read heavy work to limit slop, albeit with sub-agents and workflows).