Benchmark
AIME 2025
Math benchmarks measure how well models solve maths problems. Scores are percentages. Higher is better.
- Models measured
- 104
- Scores recorded
- 147
- Sources
- 2
- IQ weight
- 1.67
Score distribution
- 81.5%
- Mean
- 88%
- Median
- 19.2%
- Std dev
- 51.9%
- 10th percentile
- 99.6%
- 90th percentile
- 6.7%
- Min
This benchmark has a weight of 1.67 for chat models, measured over 104 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.5Python tools · 64K thinking · 200K context · avg@5 · contamination caveat· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 100% |
| 2 | Claude Sonnet 4.5Python tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 100% |
| 3 | Gemini 3 Pro Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 100% |
| 4 | Gemini 3 Pro PreviewGemini 3 Pro Model Card · results as of November 2025 · code execution· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 100% |
| 5 | GPT 5.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 100% |
| 6 | GPT 5.2no tools · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 100% |
| 7 | GPT-5.2 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 100% |
| 8 | GPT-5.2 Prono tools · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 100% |
| 9 | Kimi K2 Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 100% |
| 10 | Step 3.5 FlashStep 3.5 technical report · Table 5 · PaCoRe· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 99.9% |
| 11 | Step 3.5 FlashStep 3.5 technical report · Tables 12–13 · Python tool· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 99.8% |
| 12 | Claude Opus 4.6· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 99.8% |
| 13 | Gemini 3 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 99.7% |
| 14 | Gemini 3 Flashcode execution · pass@1· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 99.7% |
| 15 | GPT 5.1high· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 99.6% |
| 16 | LongCat-Flash-Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 99.6% |
| 17 | Nemotron 3 Nano 30B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 99.2% |
| 18 | GPT OSS 20bhigh· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 98.7% |
| 19 | GPT OSS 20bofficial model card Table 3 · reasoning high · with tools; Table 3 does not name the exact tool · pass@1· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 98.7% |
| 20 | Doubao Seed 2.0 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 98.3% |
| 21 | GPT OSS 120Bofficial model card Table 3 · reasoning high · with tools; Table 3 does not name the exact tool · pass@1· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 97.9% |
| 22 | Step 3.5 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 97.3% |
| 23 | Step 3.5 FlashStep 3.5 Flash official headline table · vanilla· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 97.3% |
| 24 | Mai Thinking 1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 97% |
| 25 | GPT-5.1 Codexhigh· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 96.7% |
| 26 | Sarvam 105B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 96.7% |
| 27 | Sarvam-30B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 96.7% |
| 28 | Claude Haiku 4.5Python tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 96.3% |
| 29 | Trinity Large ThinkingTrinity-Large-Thinking Model Card· Reported by Arcee AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 96.3% |
| 30 | Kimi K2.5setting not named by the source· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 96.1% |
| 31 | Kimi K2.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 96.1% |
| 32 | DeepSeek V3.2 Speciale· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 96% |
| 33 | DeepSeek V3.2 SpecialeSpeciale high-compute reasoning · temperature 1.0 · 128K context · pass@1· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 96% |
| 34 | GLM 4.7· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 95.7% |
| 35 | Claude Sonnet 4.6no tools · adaptive thinking · max effort · 10 trials· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 95.6% |
| 36 | Gemini 3 Flashno tools · pass@1· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 95.2% |
| 37 | Gemini 3 Pro PreviewGemini 3 Pro Model Card · results as of November 2025 · no tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 95% |
| 38 | GPT 5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94.6% |
| 39 | GPT 5high· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94.6% |
| 40 | Doubao Seed 1.8· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94.3% |
| 41 | MiMo V2 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94.1% |
| 42 | MiMo V2 FlashMiMo-V2-Flash final evaluation · published post-training table· Reported by Xiaomi MiMo· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 94.1% |
| 43 | GPT 5.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94% |
| 44 | GPT-5.1-Instant· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 94% |
| 45 | GLM 4.6· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 93.9% |
| 46 | Grok 3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 93.3% |
| 47 | DeepSeek Reasoner· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 93.1% |
| 48 | DeepSeek V3.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 93.1% |
| 49 | DeepSeek V3.2Thinking mode · temperature 1.0 · 128K context · custom step-by-step / boxed-answer prompt· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 93.1% |
| 50 | Seed 2.0 Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 93% |
Showing 1–50 of 147 results.
Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Data by LLM Stats