Evaluation sourcesOfficial evaluation
SimpleQA Verified
Epoch AI / 知识 · 短问题的事实准确性
On HOTAIRanked
Evidence budget4%
Upstream data09/30 06:00
Last successful sync10/01 20:05
What and how it measures
只取 Epoch 自行复跑的 mean_score，采用固定推理档位；不读取 ECI 或 Best score 聚合列。
How this evidence is used
正式计分；同题家族共用固定预算。
Evaluation 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。
| Source rank | Source model | Raw score | Representative configuration |
|---|---|---|---|
| 1 | gpt-6-astra_maxOpenAI | 75.6% | Max 推理 |
| 2 | gpt-6.1-sol_maxOpenAI | 73.9% | Max 推理 |
| 3 | gemini-3.1-pro-preview_highGoogle | 73.5% | High 推理 |
| 4 | claude-opus-5-5_maxAnthropic | 72.2% | Max 推理 |
| 5 | claude-fable-5-1_maxAnthropic | 70.8% | Max 推理 |
| 6 | claude-fable-5_xhighAnthropic | 70.7% | xHigh 推理 |
| 7 | gemini-3.8-flash_highGoogle | 69.7% | High 推理 |
| 7 | gpt-5.6-sol_maxOpenAI | 69.7% | Max 推理 |
| 9 | gemini-3.7-flash_highGoogle | 69.2% | High 推理 |
| 10 | gemini-3-flash-preview_highGoogle | 66.8% | High 推理 |
| 11 | gemini-3.6-flash_highGoogle | 66.2% | High 推理 |
| 11 | gemini-3.5-flash_highGoogle | 66.2% | High 推理 |
| 13 | gpt-5.5_xhighOpenAI | 63.0% | xHigh 推理 |
| 14 | gpt-6-sol_maxOpenAI | 60.7% | Max 推理 |
| 15 | muse-spark-1.2_xhighMeta | 60.3% | xHigh 推理 |
| 16 | claude-opus-5_maxAnthropic | 59.9% | Max 推理 |
| 17 | muse-spark-1.1Meta | 57.8% | 来源默认配置 |
| 18 | qwen3.7-maxAlibaba | 55.8% | 来源默认配置 |
| 19 | claude-opus-4-8_maxAnthropic | 53.0% | Max 推理 |
| 20 | deepseek-v4-pro-0813_maxDeepSeek | 52.9% | Max 推理 |
| 21 | qwen3.6-max-previewAlibaba | 52.0% | 来源默认配置 |
| 22 | claude-opus-4-7_xhighAnthropic | 51.7% | xHigh 推理 |
| 23 | kimi-k3_maxMoonshot AI | 50.6% | Max 推理 |
| 24 | gpt-5-2025-08-07_highOpenAI | 50.1% | High 推理 |
| 25 | o3-2025-04-16_highOpenAI | 49.4% | High 推理 |
| 27 | grok-4.6_xhighxAI | 48.9% | xHigh 推理 |
| 28 | qwen3-max-2025-09-23Alibaba | 48.7% | 来源默认配置 |
| 29 | grok-4.5_highxAI | 48.3% | High 推理 |
| 30 | gpt-5.1-2025-11-13_highOpenAI | 48.0% | High 推理 |
| 31 | qwen3.8-max-0902_xhighAlibaba | 47.3% | xHigh 推理 |
Evaluation limits and attribution
正式计分；同题家族共用固定预算。
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 HOTAI 共识分使用不同尺度，不能直接相加。