Skip to content

Qwen3-Max-Thinking

Released 2025-11 · reasoning · 256K tokens · 5 benchmarks · Open weight

Editorial notes

Trillion-param MoE de razonamiento. AIME 2025 100% **requiere tool use + scaled test-time compute** (sin tools no llega a 100). Removidos del catalog: Arena-ELO 1170 (era GDPval-AA mal etiquetado), LiveBench 79.3 (era del modelo base), IFEval 71.0 (era IFBench, benchmark distinto).

Spec sheet

Company
Alibaba
Country
CN
Type
reasoning
Release
2025-11
Context
256K tokens
License
qwen-license
Quants
BF16, Q8_0, Q5_K_M, Q4_K_M
Slug
qwen3-max-thinking

Benchmarks (5)

Cite this model
BibTeX · APA

BibTeX

@misc{frontier-qwen3-max-thinking,
  title  = {Qwen3-Max-Thinking},
  author = {{Alibaba}},
  year   = {2025},
  note   = {Frontier Benchmarks AI atlas. Accessed 2026-05-08},
  url    = {https://frontierbenchmarks.com/models/qwen3-max-thinking}
}

APA

Alibaba (2025). Qwen3-Max-Thinking [Large language model]. Frontier Benchmarks AI. Retrieved 2026-05-08, from https://frontierbenchmarks.com/models/qwen3-max-thinking

Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.