OpenCompass Agent - Multi-Turn (CompassBench 2601): leaderboard
OpenCompass CompassBench 2601 edition (a fresh private question set per edition), multi-turn tool-calling subset of the agent category. Score 0-100; higher is better. Pinned to the 26-01 edition because the newest CompassBench edition no longer publishes this table.
Metric: Score (%). Source: rank.opencompass.org.cn. Saturation forecast: Around February 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 (Thinking) | 84.8 | #87 (Claude Opus 4.5) |
| 2 | Gemini 3 Pro (Preview) | 82.5 | #71 |
| 3 | GLM-4.7 | 81.8 | #202 |
| 4 | Claude Sonnet 4.5 (Thinking) | 81.2 | #148 (Claude Sonnet 4.5) |
| 5 | Grok 4.1 | 80.5 | #239 |
| 6 | DeepSeek V3.2 | 78.3 | #210 |
| 7 | Kimi K2 (Thinking) | 77 | #252 (Kimi K2) |
| 8 | ERNIE 5.0 | 75.2 | #243 |
| 9 | Qwen 3 Max (2026-01-23) | 75.2 | #158 |
| 10 | GPT-5.2 (2025-12-11) (High) | 72.5 | |
| 11 | Kimi K2.5 | 72 | #152 |
| 12 | Tencent HY 2.0 Think | 72 | #206 |
| 13 | Doubao-Seed-1-8-251228 | 72 | #153 |
| 14 | MiniMax-M2.1 | 70.5 | #299 |
| 15 | O3 (2025-04-16) (High) | 66.2 | #125 (O3 (2025-04-16)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=opencompass-agent-multi-turn-compassbench-2601 · How It Works · Data refreshed daily, snapshot 2026-11-05.