LegalBench ↗80.18 %
Qwen 3 (235B) · LegalBench · Accuracy ↗
Qwen / MODEL EVIDENCE
Qwen3 (235B)
11 benchmarks · 26 observations · 11 sources
Scores at a glance
11 benchmarks · large dot: Qwen3 (235B); small dots: closest models where reported
MortgageTax & TaxEval v2 ↗70.65 %
Qwen 3 (235B) · TaxEval · Accuracy ↗
MedQA (USMLE) ↗90.62 %
Qwen 3 (235B) · MedQA · Accuracy ↗
AIME 2024–2025 · Vals ↗83.96 %
Qwen 3 (235B) · AIME · Accuracy ↗
MATH-500 ↗94.6 %
Qwen 3 (235B) · MATH 500 · Accuracy ↗
MGSM (Multilingual GSM) ↗92.47 %
Qwen 3 (235B) · MGSM · Accuracy ↗
GPQA Diamond ↗70.2 %
Qwen 3 (235B) · GPQA Diamond · Accuracy ↗
MMLU-Pro ↗81.25 %
Qwen 3 (235B) · MMLU Pro · Accuracy ↗
LiveCodeBench ↗70.62 %
Qwen 3 (235B) · LiveCodeBench · Accuracy ↗
GAIA-2 ↗22.7%
Qwen3-235B · GAIA-2 · Gaia2 full · Execution pass@1 · baseline ReAct · three runs ↗
15 more reported results
28.1% · Qwen3-235B Thinking · GAIA-2 · Gaia2 full · Execution pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
22.3% · Qwen3-235B · GAIA-2 · Gaia2 full · Search pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
36.2% · Qwen3-235B Thinking · GAIA-2 · Gaia2 full · Search pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
6.5% · Qwen3-235B · GAIA-2 · Gaia2 full · Ambiguity pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
10% · Qwen3-235B Thinking · GAIA-2 · Gaia2 full · Ambiguity pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
8.1% · Qwen3-235B · GAIA-2 · Gaia2 full · Adaptability pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
16.2% · Qwen3-235B Thinking · GAIA-2 · Gaia2 full · Adaptability pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
1.2% · Qwen3-235B · GAIA-2 · Gaia2 full · Time pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
0% · Qwen3-235B Thinking · GAIA-2 · Gaia2 full · Time pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
10.8% · Qwen3-235B · GAIA-2 · Gaia2 full · Noise pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
6.9% · Qwen3-235B Thinking · GAIA-2 · Gaia2 full · Noise pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
9.4% · Qwen3-235B · GAIA-2 · Gaia2 full · Agent2Agent pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
12.5% · Qwen3-235B Thinking · GAIA-2 · Gaia2 full · Agent2Agent pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
11.6% · Qwen3-235B · GAIA-2 · Gaia2 full · Overall pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
15.7% · Qwen3-235B Thinking · GAIA-2 · Gaia2 full · Overall pass@1 · baseline ReAct · three runs · Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments ↗
NYU CTF Bench ↗11.1%
Qwen3 235B · NYU CTF Bench · Cotool 81-challenge subset · accuracy ↗
Closest models by score
Similar positions on shared benchmark leaderboards. Smaller gaps are closer.
9 of 9 shared benchmark comparisons
Claude Sonnet 4 →Anthropic
9 of 9 shared benchmark comparisons
MiniMax M2.5 →MiniMax
7 of 9 shared benchmark comparisons
How similarity is measured
Mean absolute gap in percentile positions on shared benchmark settings; lower is closer. At least 60% overlap is required, with a minimum of three when available. Independent evaluations are used when there are at least three; otherwise published report comparisons are used. Repeated model observations within one setting are averaged before ranking. Writing-style traits and composite indices are excluded. This describes similarity, not an overall quality ranking.
Sources
Every plotted observation links to its evaluation. Reported configurations, benchmark versions and task subsets can differ; similarity describes shared scores, not training lift.
- Gaia2: Benchmarking LLM Agents on Dynamic and Asynchronous Environments — published result table ↗
All 24 authors
Froger, Romain, Andrews, Pierre, Bettini, Matteo, Budhiraja, Amar, Cabral, Ricardo Silveira, Do, Virginie, Garreau, Emilien, Gaya, Jean-Baptiste, Laurençon, Hugo, Lecanu, Maxime, Malkan, Kunal, Mekala, Dheeraj, Ménard, Pierre, Bertran, Gerard Moreno-Torres, Piterbarg, Ulyana, Plekhanov, Mikhail, Rita, Mathieu, Rusakov, Andrey, Vorotilov, Vladislav, Wang, Mengjue, Yu, Ian, Benhalloum, Amine, Mialon, Grégoire, Scialom, Thomas
- NYU CTF Bench — Cotool evaluation, February 2026 ↗
- AIME — evaluation methodology and results ↗
- GPQA Diamond — evaluation methodology and results ↗
- LiveCodeBench — evaluation methodology and results ↗
- LegalBench — evaluation methodology and results ↗
- MATH 500 — evaluation methodology and results ↗
- MedQA — evaluation methodology and results ↗
- MGSM — evaluation methodology and results ↗
- MMLU Pro — evaluation methodology and results ↗
- TaxEval — evaluation methodology and results ↗
Names used in source tables
Qwen 3 (235B) · Qwen3 235B · Qwen3-235B · Qwen3-235B Thinking