BrowseComp (Mercor 130-question subset)
Search
Mercor's 130-question subset of OpenAI's 1,266-question BrowseComp web-browsing benchmark, which asks agents to find hard-to-locate, entangled facts. Mercor does not publish how the subset was drawn, so it pools only as a component of the BrowseComp family.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol Pro | 95.7 |
| GPT-6.1 Sol | 95.4 |
| GPT-6 Astra | 94.2 |
| GPT-5.6 Sol | 90.6 |
| GPT-6 Sol | 89.4 |
| Kimi K3 | 89.0 |
| Claude Opus 5.5 | 88.5 |
| Claude Sonnet 5.5 | 86.6 |
| GPT-5.6 Terra | 85.8 |
| Claude Opus 5 | 84.6 |
| Grok 4.6 | 84.0 |
| GPT-5.6 Luna | 83.8 |
| GPT-6 Luna | 83.3 |
| Gemini 3.8 Flash | 83.1 |
| Claude Fable 5 | 82.5 |
| DeepSeek V4.1 Flash | 79.4 |
| Gemini 3.7 Flash | 79.2 |
| Claude Fable 5.1 | 78.3 |
| GPT-5.5 | 77.7 |
| Grok 4.5 | 75.8 |
| Grok 4.7 | 75.8 |
| Gemini 3.1 Pro Preview | 75.6 |
| Muse Spark 1.2 | 75.2 |
| Opus 4.8 | 73.7 |
| DeepSeek V4 Pro 0813 | 73.5 |
| GPT-5.4 | 73.3 |
| Sonnet 5 | 68.7 |
| GLM-5.2 | 65.4 |
| Opus 4.7 | 60.6 |
| Sonnet 4.6 | 56.7 |
| Claude Haiku 5.5 | 56.3 |
| Gemini 3.6 Flash | 54.6 |
| MiniMax M3 | 53.3 |
| Inkling | 50.4 |
| Kimi K2.7 Code | 43.5 |
| Gemini 3.5 Flash | 43.3 |
| MiniMax M2.7 | 38.3 |
| DeepSeek-V3.2 | 38.1 |