models · qwen

Qwen3.5-122B-A10B

Qwen3.5 MoE 122B total / 10B active. 262K context, 129 tok/s. Arena 750. GPQA Diamond 86.6%. Strong MoE efficiency at near-70B inference cost.

Compare Verify

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	9c064915c4742a03e8d5e317e3841f521e9bf51a5cedfe3829d4e173b13555a5
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	125,086,497,008 (125.1B)
Context window	262K
Modality	text+image+video->text
Native precision	BF16
Pulls recorded here	360,013
Downloads reported upstream	523,504

Reported scores

GPQA Diamond	85.7
IFBench	75.7
Humanity's Last Exam	25.2

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 3090 ×4 devices	vllm	safetensors AWQ	152 tok/s @ 2,048	candidate	localmaxxing
RTX PRO 6000 Blackwell	llama.cpp	Q4_K_P	94.6 tok/s @ 32,768	candidate	localmaxxing
RTX PRO 6000 Blackwell	llama.cpp	Q4_K_M	88.2 tok/s @ 32,768	candidate	localmaxxing
Apple M5 Max 128GB	mlx	MLX 4bit	55.9 tok/s @ 8,256	candidate	exo-postgres
Apple M3 Ultra 96GB 80-core GPU	mlx	MLX 4bit	51.1 tok/s @ 8,256	candidate	exo-postgres
Apple M3 Ultra 96GB 60-core GPU	mlx	MLX 4bit	50.4 tok/s @ 8,256	candidate	exo-postgres
GeForce RTX 3090 ×8 devices	vllm	safetensors GPTQ Int4	49.1 tok/s @ 8,192	candidate	localmaxxing
GeForce RTX 3090 ×8 devices	vllm	safetensors FP8	7.4 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
GeForce RTX 3090 ×4 devices	llama.cpp	UD-Q4_K_XL	51.3 tok/s @ 65,536	candidate	localmaxxing
NVIDIA DGX Spark GB10	vllm	INT4	33.9 tok/s @ 131,072	candidate	localmaxxing

From local-ai-registry (MIT), commit 124959d: 25 runs across 8 machines, 0 validated — model revision and runtime pinned, launch accepted — and 25 candidate, their label for useful evidence without a reproducible-launch promise. 3 more not listed.

Measured by local.ai

Hardware	Build measured	Decode @ 8K	Memory @ 8K	Decode mode
MacBook Pro M5 Max 128GB · 40-core GPU	4bit	54.9 tok/s	72.6 GB	not stated
Mac Studio M3 Ultra 96GB · 60-core GPU	4bit	50.8 tok/s	72.6 GB	not stated

Measured by local.ai, not by us: 2 runs. 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 2026-06-11. 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.