benchmark sourcesOfficial benchmark
OTIS Mock AIME 2024–2025
This page's content is in Chinese.
Epoch AI / 推理 · 高中数学竞赛难度的模拟 AIME 题
In AIDC热点Ranked
Evidence budget1.2%
Upstream data time11/03 04:57
Last successful sync11/04 14:05
What it measures and how
只取 Epoch 自行复跑的 mean_score,保留推理档位与标准误。题目来自 OTIS 模拟 AIME 2024–2025,答案为整数,可自动判分。
How to use this evidence
正式计分(方法 v16 起);与 FrontierMath 同属数学,但难度和题型不同,单列一个题目家族。
Benchmark results
按固定规则,每个公开模型采用一套代表配置。匿名测试型号不展示,保留原榜名次,每项最多 30 个。
| Original board rank | Original board model | Raw result | Representative config |
|---|---|---|---|
| 1 | gpt-5.6-sol_maxOpenAI | 100.0% | Max 推理 |
| 1 | gpt-6-astra_maxOpenAI | 100.0% | Max 推理 |
| 1 | claude-sonnet-5-5_maxAnthropic | 100.0% | Max 推理 |
| 1 | gpt-6.1-sol_maxOpenAI | 100.0% | Max 推理 |
| 1 | claude-fable-5-1_maxAnthropic | 100.0% | Max 推理 |
| 1 | claude-opus-5-5_maxAnthropic | 100.0% | Max 推理 |
| 1 | gpt-6.1-terra_maxOpenAI | 100.0% | Max 推理 |
| 1 | gpt-6-sol_maxOpenAI | 100.0% | Max 推理 |
| 1 | gemini-4-argon_highGoogle | 100.0% | High 推理 |
| 1 | qwen3.8-max-0902_xhighAlibaba | 100.0% | xHigh 推理 |
| 14 | claude-fable-5_maxAnthropic | 99.7% | Max 推理 |
| 14 | gpt-5.6-terra_maxOpenAI | 99.7% | Max 推理 |
| 16 | muse-wonder-1.0_maxMeta | 99.4% | Max 推理 |
| 16 | qwen3.8-max_xhighAlibaba | 99.4% | xHigh 推理 |
| 16 | mistral-large-4-0-oss_highMistral AI | 99.4% | High 推理 |
| 16 | deepseek-v4.1-flash-1017_maxDeepSeek | 99.4% | Max 推理 |
| 20 | grok-4.6_xhighxAI | 99.2% | xHigh 推理 |
| 20 | doubao-seed-evolving_highByteDance | 99.2% | High 推理 |
| 20 | muse-spark-1.3_xhighMeta | 99.2% | xHigh 推理 |
| 23 | gemini-3.8-flash_highGoogle | 98.9% | High 推理 |
| 23 | claude-opus-5_maxAnthropic | 98.9% | Max 推理 |
| 23 | gpt-6-luna_maxOpenAI | 98.9% | Max 推理 |
| 23 | claude-haiku-5-5_xhighAnthropic | 98.9% | xHigh 推理 |
| 27 | deepseek-v4-pro-0813_maxDeepSeek | 98.6% | Max 推理 |
| 28 | deepseek-v4.1-flash_maxDeepSeek | 98.3% | Max 推理 |
| 28 | mistral-large-4-0_highMistral AI | 98.3% | High 推理 |
| 28 | gpt-5.6-luna_maxOpenAI | 98.3% | Max 推理 |
| 28 | claude-opus-4-8_maxAnthropic | 98.3% | Max 推理 |
| 32 | Muse-Spark-Open-816B-A42B_xhighMeta | 98.1% | xHigh 推理 |
| 32 | grok-4.7_xhighxAI | 98.1% | xHigh 推理 |
Benchmark limits and data attribution
竞赛题已有大量公开训练材料,前沿模型分数接近上限,主要区分中上游模型。
Data license:CC BY 4.0 · Epoch AI 自行评测数据
成绩由 Epoch AI 发布,原始分数与 AIDC热点 共识分使用不同尺度,不能直接相加。