models · qwen
Qwen3 flagship MoE. 235B total / 22B active parameters. Hybrid thinking/non-thinking mode, 128K context. Tops open-weight coding and math benchmarks.
| Weight file SHA-256 | ac326f3a3d0e9a8c12e98ed0f1d0f790de51c3024922532724e857d55754e9cb |
|---|---|
| Method | hf-published-sha256 |
| 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 | 235,093,634,560 (235.1B) |
| Context window | 128K |
| Modality | text->text |
| Architecture | qwen3_moe |
| Layers | 94 |
| Hidden size | 4096 |
| Attention | GQA (64 heads, 4 KV) |
| Native precision | BF16 |
| Pulls recorded here | 1,200,042 |
| Downloads reported upstream | 346,403 |
| MMLU-Pro | 76.2 |
|---|---|
| GPQA Diamond | 61.3 |
| AIME 2025 | 23.7 |
| MATH-500 | 90.2 |
| LiveCodeBench | 34.3 |
| IFBench | 36.6 |
| Humanity's Last Exam | 4.2 |
Reported by a third party and recorded here, not re-run by us. Source: artificialanalysis.ai/api/v2.
| Hardware | Engine | Build measured | Decode, 1 stream | Label | Reported by |
|---|---|---|---|---|---|
| Apple M3 Ultra 96GB 80-core GPU | llama.cpp | GGUF UD-Q3_K_XL | 24.3 tok/s @ 8,256 | candidate | exo-postgres |
From local-ai-registry (MIT), commit 124959d: 1 run across 1 machine, 0 validated — model revision and runtime pinned, launch accepted — and 1 candidate, their label for useful evidence without a reproducible-launch promise.
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.