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.
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 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!
> 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that Chinese government fund Chinese models way more compared to what those companies receive (except llama, which is outdated but was strong foundation at its time)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.
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.
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.
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 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.
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.
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
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
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.
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.
TM is a special company in that a lot of well commected people are willong to fund MM SOLELY because having a woman leader looks well on their family office portfolio.
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.
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.
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/
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?
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).
> 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.
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.
Very nice, multi modal, largest open weight model that supports audio. Would be interesting to see how good the audio capability is.
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.
America needs its own DeepSeek or Z.ai, a lot of people (myself included) root for open chinese models to win because they have no other choice.
Thinking Machines might be it.
I don't hear about them a lot but it looks like arcee.ai is aiming to be just that.
Here are some of their current open weight offerings: https://www.arcee.ai/open-source-catalog
What is the business model for an open weight model?
The same business model that Deepseek is using.
Open-source models + services.
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 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!
> 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.
> That turned into a $2 billion financing with NVIDIA doing roughly $500 million and was a complete pivot.
I suspect 2B is not enough to boostrap frontier model from the scratch (for both talent and hardware)
Jensen Huang is just trying to commoditize the complements to his GPUs.
I have a similar bet. Looks like people don't like this idea. You got downvoted a lot.
isn't that what Reflection is trying to be?
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.
Not compared against Gemma 4? That is a big omission.
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.
I do grant it's a good rule of thumb.
> 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.
For the most part it’s better than Nemotron, worse than GLM. This makes it the best American open weights model from what I can tell?
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 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.
What are the different business models for open-weight AI companies?
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.
Maybe the thesis is that
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
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
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.
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.
use open models to gain marketing/users/attention and then go closed? maybe
TM is a special company in that a lot of well commected people are willong to fund MM SOLELY because having a woman leader looks well on their family office portfolio.
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.
- inference
- RLaaS (Tinker, or the more involved FDE motion a la Reflection / Applied Compute)
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.
Looks like it can be tried at https://tinker.thinkingmachines.ai/playground
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/
My personal bet is that this model should really shine in Autoresearch NanoGPT-style speedruns because its first-class integration with Tinker
too bad we'll never know how good it is, since they used a radar plot to show its benchmark scores!
How does the radar plot prevent you from looking at just one of its axes?
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?
Give me a good 180B param model that fits snuggly on an single DGX spark and I will sing your praises.
The Artifical Analysis has a link on their homepage but it 404's :/
https://artificialanalysis.ai/models/inkling
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).
It looks like HuggingFace shows Apache-2.0 but they have AUP. How does it work together?
Happy to see an open weight model ! This has all the right ingredients for success.
Do they have an api to try the model in real envs?
You certainly cooking smth, Good Luck Mira.
why is this website ai slop
If it's ~30% bigger and not as good as GLM 5.2, why would I tinker with this model?
Maybe for the multi modal?
> 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
which cheap models have you found work best as agents?
Then why are they publishing the benchmarks which makes them look worse than GLM 5.2?
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.
Because it's still informative
being close is still impressive, especially for their first (released) model
gives me hope that the training moat is even smaller than we thought
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.
> Maybe for the multi modal?
Yeah
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.
Cool, now we just need the GPU that supports it