models · Qwen

Qwen3.5-9B

Catalogued from llm-stats Open LLM Leaderboard at Quarantine — unverified, pending attestation.

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.

Method	hf-published-sha256

Re-check this yourself → Source → Full record →

What is recorded

Publisher	Qwen
Origin	Recorded as a covered nation under 10 U.S.C. § 4872(f) — the PRC, Russia, Iran or North Korea. The registry did not record which.
Licence	Apache-2.0
Parameters	9,653,104,368 (9.65B)
Context window	262,144 tokens
Modality	text+image+video->text
Native precision	BF16
Pulls recorded here	6
Downloads reported upstream	9,468,124

Reported scores

GPQA Diamond	80.6
IFBench	66.7
Humanity's Last Exam	14.9

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 5090	llama.cpp	Q6_K	170 tok/s @ 2,048	candidate	localmaxxing
GeForce RTX 5070 Ti	llama.cpp	Q4_K_S	124 tok/s @ 8,192	candidate	localmaxxing
GeForce RTX 3090 Ti	llama.cpp	UD-Q4_K_XL	119 tok/s @ 8,192	candidate	localmaxxing
GeForce RTX 3090 ×2 devices	llama.cpp	Q4_K_S	110 tok/s @ 8,192	candidate	localmaxxing
H200 141GB	ollama	GGUF Q4_K_M	103 tok/s @ 32,768	candidate	0xsero
Radeon RX 9070 XT	llama.cpp	Q4_K_M	87.5 tok/s @ 4,096	candidate	localmaxxing
Apple M3 Ultra 96GB 80-core GPU	llama.cpp	GGUF Q4_K_M	81.6 tok/s @ 8,256	candidate	exo-postgres
GeForce RTX 5080	llama.cpp	Q8_0	79.6 tok/s @ 4,096	candidate	localmaxxing
Apple M5 Max 128GB	llama.cpp	Q4_K_M	79.3 tok/s @ 2,048	candidate	localmaxxing
Apple M5 Max 128GB	llama.cpp	GGUF Q4_K_M	78.6 tok/s @ 8,256	candidate	exo-postgres

From local-ai-registry (MIT), commit 124959d: 99 runs across 51 machines, 19 validated — model revision and runtime pinned, launch accepted — and 80 candidate, their label for useful evidence without a reproducible-launch promise. 48 more not listed. 21 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.

Measured by local.ai

Hardware	Build measured	Decode @ 8K	Memory @ 8K	Decode mode
NVIDIA GeForce RTX 5090 32GB	Q4_K_M	212 tok/s	10.3 GB	ordinary
NVIDIA RTX PRO 6000 Blackwell 96GB	Q4_K_M	204 tok/s	6.9 GB	ordinary
NVIDIA RTX PRO 6000 Blackwell 96GB	FP8	140 tok/s	90.0 GB	ordinary
NVIDIA GeForce RTX 5090 32GB	FP8	139 tok/s	28.4 GB	ordinary
NVIDIA GeForce RTX 4090 24GB	Q4_K_M	129 tok/s	8.0 GB	ordinary
NVIDIA RTX 6000 Ada 48GB	Q4_K_M	127 tok/s	14.6 GB	ordinary
Mac Studio M3 Ultra 96GB · 80-core GPU	4bit	94.0 tok/s	7.5 GB	ordinary
Mac Studio M3 Ultra 96GB · 60-core GPU	4bit	90.9 tok/s	7.5 GB	ordinary

Measured by local.ai, not by us: 133 runs, 125 more not listed. Each names its engine pinned by image digest and the exact serve command, in the full record.

Elsewhere in the registry

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

Registry record as of 2026-09-26. 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.