0

Qwen3 32B

Alibaba's mid-tier dense Qwen3 model - 32B parameters with hybrid thinking modes, the largest of the dense Qwen3 lineup.

Open
Publisher
Alibaba
Family
Qwen
Params
32B
Context
131K
Input
Output
Providers
togetherfireworksollamaself-host
License
apache-2.0
Released
28 Apr 2025
Canonical
qwenlm.github.io/blog/qwen3

Cite

Notes

Only stored in your browser.

Attribution

Attribution policy →
Context
131K
Weights
Open
downloadable
Input
$0.15
/1M tok
Output
$0.60
/1M tok
Blended
$0.26
#298/ 500/1M · 3:1
Speed
0
#106/ 499tok/s

Price & speed via Artificial Analysis · ranked across tracked models

Reasoning effortIndexPriceSpeed
Reasoning7.2$0.320 t/s
Non-reasoning7.3$0.260 t/s

Intelligence Index per reasoning-effort setting, via Artificial Analysis

Reported on 17 evals across 9 domains - best 96.1% on MATH-500 · #31 of 190

Reported eval scores

17
Simple Bench20
Simple Bench20
AIME 2024: Problems from the American Invitational Mathematics Examination80.7
MATH-50096.1
MMLU-Pro79.8
Arena-Hard53.3
Humanity's Last Exam (HLE)7.4
LiveCodeBench54.6
Aider Polyglot Benchmark40
AIME 2025: Problems from the American Invitational Mathematics Examination73
SciCode36
GPQA Diamond66.8
GPQA (Full Set)65.8
τ²-bench (Tau²-bench)29.8
IFBench36.3
Terminal-Bench (Hard)3
EQ-Bench Longform Creative Writing40.5