Model datasheet · Qwen3
Qwen3 32B
The dense flagship under 35B — the only sub-35B model with an independent Aider coding score. Trades the 30B-A3B MoE's speed for steadier quality and more graceful quantization; expect <10 tok/s on unified memory.
- Vendor
- Alibaba
- Architecture
- Dense · 32.8B
- Context
- 32,768 tokens
- License
- Apache-2.0
- Released
- 2025-04-29
- Vision
- No
→ GGUF (llama.cpp / LM Studio / Ollama) → MLX (Apple Silicon)
§1 Characteristics by quantization band
— Not yet rated · contributions welcome —
Q4–Q5 — 4–5 bit
20GB weights · e.g. Q4_K_M , Q5_K_M , mlx-4bit + KV: 2GB at 8K · 8GB at 32K · 8GB at 128Kdata checked Aug 2026
- quant-degradation-community — KLD/perplexity delta vs bf16 @ Q4_K_M · aggregated · 2026-08-30
Editorial derivation: large dense models quantize gracefully; small loss at 4-bit.
data checked Aug 2026
- quant-degradation-community — KLD/perplexity delta vs bf16 @ Q4_K_M · aggregated · 2026-08-30
Editorial derivation: small loss at 4-bit.
data checked Aug 2026
- quant-degradation-community — community consensus @ Q4_K_M · aggregated · 2026-08-30
Editorial derivation: occasional argument-format slips at 4-bit.
data checked Aug 2026
- quant-degradation-community — community consensus @ Q4_K_M · aggregated · 2026-08-30
Editorial derivation: IF is not quant-sensitive.
— Not yet rated · contributions welcome —
FP16 — full precision (fp16/bf16)
66GB weights · e.g. bf16 + KV: 2GB at 8K · 8GB at 32K · 8GB at 128Kdata checked Aug 2026
- vendor-model-card — AIME'24 (thinking) = 81.4 @ bf16 · vendor-reported · 2026-08-30
MATH-500: 97.2 thinking; Codeforces rating 1977.
data checked Aug 2026
- vendor-model-card — LiveCodeBench v5 (thinking) = 65.7 @ bf16 · vendor-reported · 2026-08-30
- aider-polyglot — percent correct (diff format) = 40 @ bf16 · aggregated · 2026-08-30
The only sub-35B model on the Aider polyglot board — independent real-world editing evidence.
data checked Aug 2026
- vendor-model-card — BFCL v3 (thinking) = 70.3 @ bf16 · vendor-reported · 2026-08-30
data checked Aug 2026
- vendor-model-card — IFEval strict prompt (thinking) = 85 @ bf16 · vendor-reported · 2026-08-30
Arena-Hard: 93.8.
data checked Aug 2026
- vectara-hallucination — HHEM hallucination rate % (lower is better) = 5.9 @ bf16 · aggregated · 2026-08-30
data checked Aug 2026
- vendor-model-card — RULER avg (non-thinking) = 93.7 @ bf16 · vendor-reported · 2026-08-30
91.0 thinking. Native 32K; 128K via YaRN.
§2 Known issues & what fixes them
Thinking mode burns thousands of tokens per answer — painful at the <10 tok/s this dense 32B manages on consumer hardware, especially in agent loops.
Set enable_thinking=False or append /no_think for tasks that don't need deep reasoning; consider Qwen3-30B-A3B for ~5× the speed at similar quality.
Evidence · 2 sources · community-consensus
- NVIDIA docs on Qwen3 thinking-budget control (vendor, 2026-08-30)
- Qwen3-30B-A3B vs 32B speed/quality analysis (blog, 2026-08-30)
Endless repetition under greedy decoding — the model card warns against it explicitly.
Card samplers (0.6/0.95/20 thinking; 0.7/0.8/20 non-thinking); presence_penalty up to 1.5.
Evidence · 2 sources · community-consensus
- Qwen3-32B card (greedy-decoding warning) (vendor, 2026-08-30)
- llama.cpp issue: Qwen3 GGUF repetition (github, 2026-08-30)