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

If I use this model, will I get paid by Nvidia to buy graphics cards?

I suggested to my local restaurant to pay me for eating there and pointed out it was a win-win situation because it would increase the restaurant's ARR. I was kicked out.

Of course I made a blog post afterwards about the restaurant being negative luddites and backwards.

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

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-...

Nice cadence of releases by the Nemotron team :)