I don't agree with that conclusion. I'm looking at the benchmarks page for LFM2.5-24B-A2B, and it appears to get a worse score on SWE-Bench-Verified (by a lot) when compared to Qwen 3.6 35B-A3B. It only clearly beats Qwen 3.6 35B on AutomationBench, ToolSandbox, and LiveClaw.

At least for my use cases, Qwen 3.6 35B-A3B appears better than this new release. An MoE for Qwen 3.8 would be even better than that!

▲

selcuka 6 hours ago | parent | next [-]

Would you call Qwen 3.6 35B "similar size" with ~50% more parameters (both active and total) though? I believe what GP meant that if there was a Qwen 3.8 24B-A2B, it would roughly be as intelligent as this one.

▲

onlyrealcuzzo 6 hours ago | parent | prev [-]

Way more excited for Nvidia's A9B model coming out tomorrow.

▲

cmrdporcupine 6 hours ago | parent [-]

Why? Their models are useful as a starting place for doing your own fine tuning, but they have never stood well on their own.

▲

htrp 6 hours ago | parent [-]

they always undertrain the model

▲

cmrdporcupine 6 hours ago | parent [-]

yeah it's kind of the point. they're pedagogical tools. they give you the full recipe for building it from scratch or fine-tuning it beyond what it is.

but i'm prepared to be surprised, maybe they'll conjure up something standalone useful this time

▲

dofm 5 hours ago | parent [-]

They have an incentive to produce a more finished model to go with the launch of the RTX Spark laptops, I guess.

▲

nl 4 hours ago | parent [-]

Not really - you pay for the hardware. Everyone else already is building models for Spark hardware.

	▲	dofm 3 hours ago \| parent [-]
		I think their entire pitch about local AI is best demonstrated with their own model that they can optimise for it, and the shape of this model seems optimised for the smaller RTX chips. Indeed, this model appears to be part of their launch picture: https://blogs.nvidia.com/blog/local-ai-rtx-spark-laptops-sof...