Nvidia Nemotron 3.5 Lightning

This links to the nvfp4 version of the model, so they only compare it to the bf16 in this model card. If you're looking for other similar model comparisons they are in the model card of the bf16 version here [1].

While it looks "behind" the qwen equivalent model on most benchmarks, a few personal notes:

- nemotron models feel to me a bit less benchmaxxed / "stubborn". That means that they generalise a bit better, or can be tasked to solve similar but not quite identical task types to the training data (something that's hard to do w/ qwen/ds models)

- nemotron series are also open training (w/ open training recipes and some training data public)

- nvda will have an incentive to continue this kind of releases, even if other parties slowly abandon the open release of models. Whatever other incentives 3rd party labs have (i.e. meta, goog w/ gemma, the chinese labs that IPOd, etc) nvda will always want to sell hardware so their incentive to keep pushing open models is evident and will likely continue "forever".

[1] - https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-...

Crazy to see how well Qwen3.6 35b-a3b is holding up, sure it is ~20% larger but it's scores are also ~20% higher with the same number of active params (excluding the IFBench).

Hopefully Qwen follows up their 3.8 launch with a new 35b-a3b

Runs fast on my Mac - ~100 tokens/second - but it's a bit of an over-thinker. It sketched out four different SVGs in the reasoning trace before returning a (bad) pelican riding a bicycle: https://tools.simonwillison.net/markdown-svg-renderer#url=ht...

I used LM Studio and this model file: https://lmstudio.ai/models/nvidia/nemotron-3.5-lightning

I got a pretty solid (and reasonably fast) result for this simply coding agent activity (with simonw/llm-coding-agent):

  llm code -m lmstudio/nvidia/nemotron-3.5-lightning \
    'how does auth work?'

Developing on the Mamba 2 architecture is a really interesting point to note. It seems to be catching up to “regular” transformer architectures.

I know very little about this but why isn't it mamba-3? Isn't it... uh, more gooder?

You might look at this and and be a bit disappointed by the performance against qwen and gemma models - but this is an entirely open source training pipeline, this is quite impressive and I don't think another model this performant exists with fully open source data and recipes alongside the weights.

Interesting to see in the Agentic Coding Benchmarks that the codex harness is so demonstrably lower than any of the other harnesses, even Claude Code another lab harness. Does anyone know why that would be?

That SWEBench drop going from bf16 to fp4 is massive

Nice cadence of releases by the Nemotron team :)

[flagged]

Nvidia just throwing something "for peasants" to stay relevant. Where is competition spirit? More importantly why Nvidia is gatekeeping computing for everyday people?

I find these releases are bad taste.

Make 1TB DGX priced affordably, not some crap model for people to waste time on.