- MMLU-Pro 85.7
- MMLU-Redux 92.8
- C-Eval 93.7
- GPQA 87.4
- Best of HLE / HLE (w/ tools) 49.8
- LiveCodeBench v6 85.9
- HMMT Feb 25 98
- HMMT Nov 25 94.7
- IMOAnswerBench 83.9
- SWE Verified 75.3
- IFBench 70.9
- MultiChallenge 63.3
- Arena-Hard v2 90.2
- Tau² Bench 82.1
- BFCL-V4 67.7
- Vita Bench 40.9
- Deep Planning 28.7
- AA-LCR 68.7
Qwen3-Max-2026-01-23
Qwen3-Max-2026-01-23 places in the top three on AIDABench.
Qwen3-Max-2026-01-23
Closed weights Flagship
- AA-LCR (Artificial Analysis Long Context Reasoning) 68.7% #10 /38
- ClawProBench 55.8% #38 /50
- AIDABench 56.7% #3 /8
How does it do against its rivals?
Qwen3-Max-2026-01-23 beats Kimi K2 Thinking on 6 of the 8 benchmarks they share. Seed2.0 Pro does best against it, ahead on 6 of 8.
- Kimi K2 Thinking 6 ahead, 2 behind Compare
-
DeepSeek-V3.2 6 ahead, 3 behind Compare
-
Step 3.5 Flash 5 ahead, 4 behind Compare
- GLM-5 3 ahead, 5 behind Compare
- Kimi K2.5 3 ahead, 5 behind Compare
-
Seed2.0 Pro 2 ahead, 6 behind Compare
Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards
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
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Extended NYT Connections | 30.1% | #86 of 109 | Official leaderboard | |
| AA-LCR (Artificial Analysis Long Context Reasoning) | 68.7% | #10 of 38 lab numbers | Lab model card | |
| MMLU-Redux | 92.8% | #17 of 45 lab numbers | Lab model card | |
| Arena-Hard v2.0 | 90.2% | Not ranked | Lab model card | |
| C-Eval | 93.7% | Not ranked | Lab model card | |
| Humanity's Last Exam | 49.8% | Not ranked | Lab model card | |
| IFBench | 70.9% | Not ranked | Lab model card | |
| MMLU-Pro | 85.7% | Not ranked | Lab model card | |
| MultiChallenge | 63.3% | Not ranked | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 87.4% | Not ranked | Lab model card | |
| HMMT 2025 | 98% | Not ranked | Lab model card | |
| HMMT Nov 2025 | 94.7% | Not ranked | Lab model card | |
| IMO-AnswerBench | 83.9% | Not ranked | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-CI | 32% | #14 of 16 | Official leaderboard | |
| LiveCodeBench v6 | 85.9% | Not ranked | Lab model card | |
| SWE-bench Verified | 75.3% | Not ranked | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ClawProBench | 55.8% | #38 of 50 | Official leaderboard | |
| BFCL V4 | 67.7% | Not ranked | Lab model card | |
| DeepPlanning | 28.7% | Not ranked | Lab model card | |
| tau2-bench | 82.1% | Not ranked | Lab model card | |
| VITA-Bench | 40.9% | Not ranked | Lab model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AIDABench | 56.7% | #3 of 8 | Official leaderboard |
What does it cost to run?
On the one leaderboard that publishes a price, its best run costs 2.2 times the median model's. There, another model comes close to its score for less.
| Benchmark | Score | Price | From cheapest to priciest | Against the median | Close to its score for less Close for less |
|---|---|---|---|---|---|
| ClawProBench | 55.8% | $41.29 per run 2.2× the median | 2.2× | |
Its lab's numbers put it at 150.3, number 95 of 412
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 147.8 to 152.7. How we compute it.
- HMMT 2025 98
- HMMT Nov 2025 94.7
- C-Eval 93.7
- MMLU-Redux 92.8
- Arena-Hard v2.0 90.2
- GPQA Diamond 87.4
- LiveCodeBench v6 85.9
- MMLU-Pro 85.7
- IMO-AnswerBench 83.9
- tau2-bench 82.1
- SWE-bench Verified 75.3
- IFBench 70.9
- AA-LCR (Artificial Analysis Long Context Reasoning) 68.7
- BFCL V4 67.7
- MultiChallenge 63.3
- Humanity's Last Exam 49.8
- VITA-Bench 40.9
- DeepPlanning 28.7
Who measured it?
Independent runs cover 4 of its 22 benchmarks. For the other 18, we have only Qwen's own number or a claim Epoch AI collected.
- An independent run only 4
- A claim only 18
Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.