BrowseComp — Named context-management harness, distinct tools axis; MiniMax-sourced single-vendor scaffold
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Qwen3.5 397B A17B | Alibaba | 78.6 | 1 | 2026-08-24 |
| 2 | Inkling-Small | Thinking Machines | 77.1 | 1 | 2026-08-24 |
| 3 | MiniMax M2.7 | MiniMax | 76.3 | 1 | 2026-08-24 |
| 4 | Kimi K2.5 | Moonshot | 74.9 | 1 | 2026-08-24 |
| 5 | DeepSeek-V4-Flash | DeepSeek | 73.2 | 1 | 2026-08-24 |
| 6 | GPT-5.2 | OpenAI | 70 | 2 | 2026-08-24 |
| 7 | Gemini 3 Pro | 67.6 | 2 | 2026-08-24 | |
| 8 | DeepSeek-V3.2 | DeepSeek | 67.6 | 1 | 2026-06-05 |
| 9 | MiniMax M2.1 | MiniMax | 62 | 1 | 2026-06-05 |
| 10 | Claude Opus 4.5 | Anthropic | 59.2 | 2 | 2026-08-24 |
| 11 | MiniMax M2 | MiniMax | 56.9 | 1 | 2026-06-05 |
| 12 | Claude Sonnet 4.5 | Anthropic | 26.1 | 1 | 2026-06-05 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.