models · qwen

Qwen3-32B

Qwen3 32B dense model with hybrid thinking/non-thinking mode. Strong coding and math.

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	dfbadf618f9612482bbed4bee379953b7094fc979770e5b1f8d55038f7457b1a
Method	hf-published-sha256

Re-check this yourself → Source → Full record →

What is recorded

Publisher	qwen
Origin	China
Licence	Apache-2.0
Parameters	32,762,123,264 (32.8B)
Context window	131,072 tokens
Modality	text->text
Architecture	qwen3
Layers	64
Hidden size	5120
Attention	GQA (64 heads, 8 KV)
Native precision	BF16
Pulls recorded here	94,021
Downloads reported upstream	4,674,951

Reported scores

MMLU-Pro	72.7
GPQA Diamond	53.5
AIME 2025	19.7
MATH-500	86.9
LiveCodeBench	28.8
IFBench	31.5
Humanity's Last Exam	4.1

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
Apple M1 Max 64GB	ollama	Q4_K_M	9.0 tok/s @ 2,048	candidate	localmaxxing

From local-ai-registry (MIT), commit 124959d: 2 runs across 2 machines, 0 validated — model revision and runtime pinned, launch accepted — and 2 candidate, their label for useful evidence without a reproducible-launch promise.

Measured by local.ai

Hardware	Build measured	Decode @ 8K	Memory @ 8K	Decode mode
NVIDIA RTX PRO 6000 Blackwell 96GB	—	22.4 tok/s	89.9 GB	not stated
NVIDIA DGX Spark 128GB	—	3.5 tok/s	113.4 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.