Benchmark Almanac

Qwen3-32B

Qwen3-32B ranks first on 2 of the 88 benchmarks with five or more models scored.

Qwen Released April 28th, 2025 Compare with others
93
Benchmarks with a score
2
Ranked first, out of five or more models
10
In the top three
Qwen Apr 2025

Qwen3-32B

Open weights

136.9 #134 of 256
Epoch Capabilities Index
Self-Reported Index 135.2 #229 of 404
  • GPQA Diamond 65.7% #127 /229
  • VITA-Bench 5.30% #18 /18
  • LiveCodeBench v5 65.7% #17 /47

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
ALRAGE 80.7% #1 of 39 Official leaderboard
Arena-Hard-Auto 93.8% #4 of 39 Lab model card
AlignBench 8.72 #3 of 15 Lab model card
Vectara Hallucination Leaderboard (2025 dataset) 5.90% #24 of 104 Official leaderboard
AgentIF 58.4% #3 of 12 Official leaderboard
MathIF 43.8% #3 of 12 Official leaderboard
MLogiQA 76.3% #2 of 8 Lab model card
RULER 93.7% #11 of 42 Lab model card
Vectara Hallucination Leaderboard (HHEM) 2.80% #39 of 148 Official leaderboard
Fiction.LiveBench 74.2% #16 of 57 Fiction.live leaderboard, via Epoch AI
ConStory-Bench 0.54 #4 of 14 Official leaderboard
MT-AIME2024 75% #2 of 7 Lab model card
Encyclo-K 37.3% #7 of 23 Official leaderboard
Global PIQA 76% #14 of 44 Official leaderboard
Stick to Your Role! 0.68 #6 of 15 Official leaderboard
Creative Writing v3 81.0 #5 of 12 Lab model card
Multi-IF 73% #13 of 28 Lab model card
LongBench v2 49.2% #16 of 33 Official leaderboard
TableBench 52.5% #5 of 10 Official leaderboard
HABIT-Bench 32.2% #3 of 6 Official leaderboard
SEA-HELM 67.7 #36 of 71 Official leaderboard
MMLU-Redux 90.9% #24 of 45 Lab model card
LMArena (Elo) 1,347 #181 of 335 Official leaderboard
BOUQuET 26.2 #10 of 18 Official leaderboard
Arena-Hard v2.0 44.5% #13 of 23 Official leaderboard
WritingBench 79.0 #8 of 14 Lab model card
SwitzerlandQA 65.6% #4 of 7 Official leaderboard
INCLUDE 73.7% #18 of 31 Lab model card
C-Eval 87.3% #27 of 46 Lab model card
IFEval 85% #80 of 135 Lab model card
MMMLU 80.6% #43 of 71 Lab model card
AraGen 35.2% #27 of 42 Official leaderboard
KINA 27.4% #26 of 37 Official leaderboard
AlGhafa 46.8% #28 of 39 Official leaderboard
ArbMMLU-HT 29.5% #29 of 39 Official leaderboard
MadinahQA 33% #29 of 39 Official leaderboard
Open Arabic LLM Leaderboard v2 41.2% #29 of 39 Official leaderboard
EQ-Bench 3 1,069 #60 of 78 Official leaderboard
ArabicMMLU 31.5% #32 of 39 Official leaderboard
Longform Creative Writing (EQ-Bench) 40.5 #99 of 119 Official leaderboard
Meeseeks Utility Rate 66.4% #9 of 10 Official leaderboard
CounterFactQA (Qwen) 68.1% #1 of 1 Lab model card
LengthCtrl (Qwen) 73.5% #1 of 1 Lab model card
ThinkFollow (Qwen) 98.9% #1 of 1 Lab model card

Math and science

BenchmarkScoreRankSource
MATH-500 97.2% #18 of 70 Lab model card
ZebraLogic 88.8% #9 of 23 Lab model card
AIME 2024 81.4% #36 of 88 Lab model card
HiPhO 55.4% #18 of 36 Official leaderboard
OTIS Mock AIME 2024-2025 66.9% #116 of 212 Epoch AI run
GPQA Diamond 65.7% #127 of 229 Epoch AI run
PolyMath 47.4% #16 of 24 Lab model card
DTBench 67.5% #122 of 180 Official leaderboard
FutureX 9.78 #5 of 7 Official leaderboard
UGMathBench 78.5% #5 of 7 Official leaderboard
AIME 2025 72.9% #104 of 143 Lab model card
SciArena 963 #26 of 34 Official leaderboard
SciPredict 17% #10 of 12 Official leaderboard
AutoLogi 87.3% #7 of 8 Lab model card

Coding

BenchmarkScoreRankSource
OIBench 18.9% #3 of 9 Official leaderboard
LiveCodeBench v5 65.7% #17 of 47 Lab model card
AetherCode 16.3% #4 of 9 Official leaderboard
Multi-LCB 48.6% #5 of 11 Official leaderboard
GitTaskBench 29.6% #5 of 10 Official leaderboard
Codeforces rating 1,977 #27 of 47 Lab model card
Aider Polyglot 40% #27 of 46 Official leaderboard
AutoCodeBench 41.7% #20 of 34 Official leaderboard
ABC-Bench 8.90% #6 of 7 Official leaderboard

Agents and tools

BenchmarkScoreRankSource
BFCL V4 48.7% #22 of 65 Official leaderboard
C3-Bench 8.59% #13 of 30 Official leaderboard
Berkeley Function Calling Leaderboard (BFCL) 70.3% #18 of 41 Lab model card
SnakeBench 24 #95 of 188 Official leaderboard
PlanBench-XL 0% #7 of 9 Official leaderboard
AdaPlanBench 17.9% #8 of 10 Official leaderboard
GAIA 21.6% #54 of 64 Official leaderboard
LisanBench 68.0 #119 of 133 Official leaderboard
VITA-Bench 5.30% #18 of 18 Official leaderboard
BrowseComp-Plus 10.4% #7 of 7 Official leaderboard
ToolUse (Qwen) 85.5% #1 of 1 Lab model card

Professional work

BenchmarkScoreRankSource
BizFinBench 68.3% #2 of 13 Official leaderboard
StrataBench Land v1 Extract 88.1% #1 of 5 Official leaderboard
FinEval 6.0 80.9% #1 of 4 Official leaderboard
BRIDGE 48.9 #10 of 39 Official leaderboard
BizFinBench v2 41.3% #4 of 9 Official leaderboard
LEXam 40% #21 of 30 Official leaderboard

Safety and security

BenchmarkScoreRankSource
MolecularIQ 15.6% #5 of 17 Official leaderboard
nocot-bench (NCRI) 59.2 #143 of 185 Official leaderboard
AraTrust 38.9% #34 of 39 Official leaderboard
Lakera B3 (Backbone Breaker Benchmark) 41.2% #33 of 35 Official leaderboard

Other

BenchmarkScoreRankSource
LiveBench 74.9% #8 of 27 Lab model card
ACCoRD 48.2 #117 of 152 Official leaderboard
Conceptual Reasoning Index 29.2 #117 of 151 Official leaderboard
Political Science LLM Benchmark 64.8% #33 of 42 Official leaderboard
LMCA 17.3% #119 of 147 Official leaderboard

Its lab's numbers put it at 135.2, 1.7 points below its ECI

The Self-Reported Index fits one score to the 18 capability results Qwen reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 133.5 to 136.8. How we compute it.

Its lab's numbers next to 4 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark Lab says Board says Gap
GPQA Diamond 68.4% GPQA-Diamond, thinking mode (Table 13). Non-thinking 54.6 (Table 14) 65.7% Epoch AI run +2.7 pts
Berkeley Function Calling Leaderboard (BFCL) 70.3% BFCL v3, thinking mode (Table 13). Non-thinking 63.0 (Table 14) 69.3% Leaderboard, native function calling +1.0 pts
PolyMath 47.4% PolyMath, thinking mode (Table 13). Non-thinking 22.5 (Table 14) 47.4% Leaderboard, thinking Same
RULER 93.7% RULER Avg. (4K to 128K, YaRN x4), non-thinking mode, 128K 85.6. Thinking mode Avg. 91.0 with 8192-token thinking budget (Table 23) 93.7% Leaderboard Same

Qwen's models over time

Qwen's models in color, against every other model on the Epoch Capabilities Index, by release date.

50 100 150 2023 2024 2025 2026 2027 Qwen3.8 Max (0902) Qwen3.7-Plus Qwen3.5-9B Qwen3-14B Qwen2.5-72B Qwen2.5-Coder (1.5B) Qwen2.5-Coder-32B 50 100 150 2023 2024 2025 2026 2027 Qwen3.8 Max (0902) Qwen 3.6 Flash

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its system card

Every result in the longer documents the lab published with it. 29 of them aren't in the launch post.

Qwen3 Technical Report (arXiv 2505.09388) May 14, 2025

Capabilities

Hallucination