models · qwen

Qwen3.5-35B-A3B

Qwen3.5 35B MoE with 3B active params. Ultra-fast inference at 233 tok/s.

What this registry has checked

Quarantine. We hold the publisher’s own published hash for these weights and have recorded it against this entry. We have not downloaded and hashed the weights ourselves, so the hash is the publisher’s claim, not our measurement.

Weight file SHA-256	e1dfb292d55fb73a30e536d1f4a3b73a2d56c151369b807ad1e0a11a321c880e
Method	hf-published-sha256

Re-check this yourself → Source → Full record →

What is recorded

Publisher	qwen
Origin	China
Licence	Apache-2.0
Parameters	35,951,822,704 (36B)
Context window	262,144 tokens
Modality	text+image+video->text
Native precision	BF16
Pulls recorded here	130,757
Downloads reported upstream	2,048,891

Reported scores

GPQA Diamond	84.5
IFBench	72.5
Humanity's Last Exam	21

Reported by a third party and recorded here, not re-run by us. Source: artificialanalysis.ai/api/v2.

Measured on real hardware

None of these runs are ours. Each figure is the reporter’s, linked to their own evidence, under their own trust label. The number shown is single-stream decode — one user, concurrency 1 — because aggregate throughput across many users is a different quantity and reads as far faster than anyone will see.

Reported to local-ai-registry

Hardware	Engine	Build measured	Decode, 1 stream	Label	Reported by
GeForce RTX 5070	llama.cpp	IQ1_M	182 tok/s @ 2,048	candidate	localmaxxing
GeForce RTX 4090	llama.cpp	UD-Q4_K_XL	157 tok/s @ 2,048	candidate	localmaxxing
GeForce RTX 3090	llama.cpp	UD-Q4_K_XL	124 tok/s @ 2,048	candidate	localmaxxing
Ryzen AI Max+ 395	llama.cpp	Q4_K_M	73.8 tok/s @ 32,768	candidate	localmaxxing
Ryzen AI Max+ 395	llama.cpp	Q4_K_M	58.3 tok/s @ 8,192	candidate	localmaxxing
Measured outside 2K–32K context — decode speed depends heavily on context, so these are not comparable with the rows above
Apple M5 Max 128GB	ollama	NVFP4	122 tok/s @ 262,144	candidate	localmaxxing
Ryzen AI Max+ 395	llama.cpp	Q4_K_M	72.7 tok/s @ 80,000	candidate	localmaxxing
Intel Arc Pro B70	llama.cpp	Q4_K_XL	62.8 tok/s @ 262,144	candidate	localmaxxing
Intel Arc Pro B70	llama.cpp	Q5_K_M	61.5 tok/s @ 262,144	candidate	localmaxxing
NVIDIA DGX Spark GB10	llama.cpp	UD-Q4_K_XL	60.3 tok/s @ 262,144	candidate	localmaxxing

From local-ai-registry (MIT), commit 124959d: 21 runs across 12 machines, 0 validated — model revision and runtime pinned, launch accepted — and 21 candidate, their label for useful evidence without a reproducible-launch promise. 2 more not listed. 1 report figures that contradict each other — prefill below decode, which parallel prefill essentially never is — and are left to the source rather than ranked here.

Elsewhere in the registry

All models → Browse with filters → What will run on my hardware? →

Registry record as of 2026-06-11. Identity, licence and parameter count are properties of a release and do not move; tier can, and the verify link above re-checks it against the live registry rather than this page.