Laguna S 2.1 is really great too, in the "preview" release they've done so far at least. Still pending some reasoning-looping, but besides that, it's a really strong model to run within 96GB VRAM with the NVFP4 variants, and it's really good at coding (specifically).
I found Laguna S to be pretty good at coding, pretty fast, but pretty bad as an agent - not proactive, would frequently stubbornly argue things that weren't true, and pretty bad general knowledge.
But as a pure coding model, pretty good.
Deepseek v4 Flash 0731 is so much better if you can run it, though.
Grain of salt, I think I grabbed Laguna after they fixed the initial looping issues, didn't notice those, but there might've been other fixes since.
Like glm-5.x I think it has enormous self introspection that it often trips up on, but that this self reflection is actually a superpower, that enables incredibly good output. And from (in some cases) very small models.
If you watch it think, which you can, unlike American closed models, you can steer it. You can provide a a massive rocket ship stratospheric boost to help it orient itself. You have no self correction, there is no multiplayer in American proprietary models.
Sure it's great having super powerful mystic oracles that have the "right" answers. But I love respect & revere the open thinking. No it's not automous. But it is brilliant. And it considers. A lot. Deeply. It chases. That to me is the most human of models, even as it falls far astray.
You should help it. You can. Unlike these vicious dark surfaces which yield and tell you nothing. I think this is the actual meta-core-super-point of "The session you cannot take with you" (link below). It's the session that does not care about you, will not interact with you, will not peer with you, that is a dead remote far off oracle to you. Fuck these "oracles". They are a plague against the human spirit. We should alloy humanity and AI to Augment Intellect (Engelbart). (To do less is species treason.)
https://earendil.com/posts/session-portability/https://news.ycombinator.com/item?id=49118781
Yeah I think it got bad press because the chat templates (or something?) were messed up on first release, but I've been using a quant of it and it's a powerhouse, better than qwen 3.6 27b for local on a 3090, which is saying a lot.
Looks like on <https://arena.ai> agent arena (grouped by lab) Nvidia is 15/15 (much worse than Thinky and Mistral) and on text arena it's 18/27
On <https://openrouter.ai/models?order=most-popular> I definitely see usage though (probably mostly cause Nemotron 3 Ultra is free) the grouped order is DeepSeek, Tencent, Xiaomi, OpenAI, Z.ai, Nvidia
I think glancing at a random snapshot from today misses all the context. Nemotron 3 is far more significant than you're giving it credit for.
At this point, Nemotron 3 is really an 8 month old model series. That's when Nemotron 3 Nano was released, and the Nemotron 3 Super/Ultra models this year are obviously based on that recipe, mostly just bigger with a few tweaks here and there. Against today's models, no, not that interesting. Each of the Nemotron 3 models were briefly competitive when they launched, but never exceptional, and less competitive with each scale up. The fact that it took so long for Nemotron 3 Ultra to launch really hampered its competitiveness.
The Nemotron 3 series is extremely open about training recipes and training data, far more open than most open weight models, and that is valuable.
Before Nemotron 3, Nvidia had never released a single LLM that I would consider interesting at all, so Nemotron 3 was a big step up. The closest thing was Mistral NeMo, but a significant part of the credit there goes to the Mistral team, not Nvidia.
Given how much Nemotron 3 improved, I'm curious to see if Nemotron 4 will take them to a leading edge level instead of just briefly competitive.
(Nvidia released a Nemotron 3 and a Nemotron 4 like 3 years ago... this year's Nemotron 3 is entirely unrelated. Nvidia's naming schemes leave a little bit to be desired.)
Laguna S 2.1 is really great too, in the "preview" release they've done so far at least. Still pending some reasoning-looping, but besides that, it's a really strong model to run within 96GB VRAM with the NVFP4 variants, and it's really good at coding (specifically).
There was an obvious problem in the original release, they re issued it after like a week with the reasoning looping supposedly fixed.
No it doesn't follow instructions and is substantially slower than ds4.
It's a lot smaller, and runs (quantized) on a 3090 quite well. Ds4 flash 0731 you're talking about? It's great but it's much harder to run locally.
I found Laguna S to be pretty good at coding, pretty fast, but pretty bad as an agent - not proactive, would frequently stubbornly argue things that weren't true, and pretty bad general knowledge.
But as a pure coding model, pretty good.
Deepseek v4 Flash 0731 is so much better if you can run it, though.
Grain of salt, I think I grabbed Laguna after they fixed the initial looping issues, didn't notice those, but there might've been other fixes since.
Are you talking about S or XS? S is too large for a 3090 at 118b parameters.
Like glm-5.x I think it has enormous self introspection that it often trips up on, but that this self reflection is actually a superpower, that enables incredibly good output. And from (in some cases) very small models.
If you watch it think, which you can, unlike American closed models, you can steer it. You can provide a a massive rocket ship stratospheric boost to help it orient itself. You have no self correction, there is no multiplayer in American proprietary models.
Sure it's great having super powerful mystic oracles that have the "right" answers. But I love respect & revere the open thinking. No it's not automous. But it is brilliant. And it considers. A lot. Deeply. It chases. That to me is the most human of models, even as it falls far astray.
You should help it. You can. Unlike these vicious dark surfaces which yield and tell you nothing. I think this is the actual meta-core-super-point of "The session you cannot take with you" (link below). It's the session that does not care about you, will not interact with you, will not peer with you, that is a dead remote far off oracle to you. Fuck these "oracles". They are a plague against the human spirit. We should alloy humanity and AI to Augment Intellect (Engelbart). (To do less is species treason.) https://earendil.com/posts/session-portability/ https://news.ycombinator.com/item?id=49118781
I like the transparency of its reasoning, and I agree with you, OpenAI/Anthropic/Google should show the reasoning traces as well.
Yeah I think it got bad press because the chat templates (or something?) were messed up on first release, but I've been using a quant of it and it's a powerhouse, better than qwen 3.6 27b for local on a 3090, which is saying a lot.
Just looked into some Nemotron stats
Looks like on <https://arena.ai> agent arena (grouped by lab) Nvidia is 15/15 (much worse than Thinky and Mistral) and on text arena it's 18/27
On <https://openrouter.ai/models?order=most-popular> I definitely see usage though (probably mostly cause Nemotron 3 Ultra is free) the grouped order is DeepSeek, Tencent, Xiaomi, OpenAI, Z.ai, Nvidia
I think glancing at a random snapshot from today misses all the context. Nemotron 3 is far more significant than you're giving it credit for.
At this point, Nemotron 3 is really an 8 month old model series. That's when Nemotron 3 Nano was released, and the Nemotron 3 Super/Ultra models this year are obviously based on that recipe, mostly just bigger with a few tweaks here and there. Against today's models, no, not that interesting. Each of the Nemotron 3 models were briefly competitive when they launched, but never exceptional, and less competitive with each scale up. The fact that it took so long for Nemotron 3 Ultra to launch really hampered its competitiveness.
The Nemotron 3 series is extremely open about training recipes and training data, far more open than most open weight models, and that is valuable.
Before Nemotron 3, Nvidia had never released a single LLM that I would consider interesting at all, so Nemotron 3 was a big step up. The closest thing was Mistral NeMo, but a significant part of the credit there goes to the Mistral team, not Nvidia.
Given how much Nemotron 3 improved, I'm curious to see if Nemotron 4 will take them to a leading edge level instead of just briefly competitive.
(Nvidia released a Nemotron 3 and a Nemotron 4 like 3 years ago... this year's Nemotron 3 is entirely unrelated. Nvidia's naming schemes leave a little bit to be desired.)
Nemotron 3 also introduced LatentMoE, which was adopted by Kimi K3 :)
Nemotron is a very nice model with an excellent license as well.
ArenaAI Agent Leaderboard direct link: https://arena.ai/leaderboard/agent
Don't forget IBM