推理能力榜
GPQA Diamond 研究生级科学题,博士撰写并交叉验证,答案无法通过搜索获得。
| # | 模型 | 数值 | 相对第一 | 综合 v5 | 输出 $ | 类型 |
|---|---|---|---|---|---|---|
| 1 | Gemini 4 ArgonGoogle | 96.4 | 第 1 | 50.4 | $10.00 | |
| 2 | GPT-6 AstraOpenAI | 95.8 | 99% | 56.9 | $50.00 | |
| 3 | Claude Sonnet 5.5Anthropic | 95.6 | 99% | 54.2 | $10.00 | |
| 4 | GPT-6.1 SolOpenAI | 95.4 | 99% | 54.5 | $10.00 | |
| 5 | Gemini 3.8 FlashGoogle | 95.4 | 99% | — | $3.75 | |
| 6 | GPT-6.1 TerraOpenAI | 94.3 | 98% | 51.0 | $5.00 | |
| 7 | Claude Opus 5.5Anthropic | 93.9 | 97% | 58.4 | $20.00 | |
| 8 | Kimi K3 开源Moonshot AI | 93.1 | 97% | — | $15.00 | 开源 |
| 9 | Grok 4.7xAI | 92.7 | 96% | 35.0 | $6.00 | |
| 10 | Qwen3.8 MaxAlibaba Qwen | 92.7 | 96% | — | $8.00 | |
| 11 | GLM-5.3 开源Z.ai | 90.9 | 94% | 32.0 | $2.20 | 开源 |
| 12 | MiniMax M3 开源MiniMax | 90.9 | 94% | — | $1.00 | 开源 |
| 13 | DeepSeek V4.1 Flash 开源DeepSeek | 89.8 | 93% | 32.5 | $1.20 | 开源 |
| 14 | Gemma 4 31B 开源Google | 75.8 | 79% | — | $0.40 | 开源 |
| 15 | gpt-oss-120b 开源OpenAI | 75.8 | 79% | — | $0.20 | 开源 |
| 16 | MiMo-V2.6 Pro 开源Xiaomi | 59.5 | 62% | 30.0 | $1.40 | 开源 |