OpenCompass Agent - Hard Multi-Turn (CompassBench 2507): leaderboard
OpenCompass CompassBench 2507 edition (file dated 2025-08-12): hard multi-turn tool calling sub-task of the agent category. CompassBench is OpenCompass's periodic evaluation of chat models on a private question set; each edition is a separate board. Edition 2510 republished 9 of these results for the same task set; those models appear on that board only. Score 0-100, higher is better.
Metric: Score (%). Source: rank.opencompass.org.cn. Saturation forecast: Estimated already saturated. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Doubao-Seed-1.6-thinking-250715 | 58 | #169 |
| 2 | Claude Sonnet 4 (Thinking) | 50 | #207 (Claude Sonnet 4) |
| 3 | Hunyuan-T1-20250711 | 50 | #277 |
| 4 | GPT-5 | 46 | #105 |
| 5 | Qwen 3 235B A22B 2507 Instruct | 44 | #315 |
| 6 | ERNIE-X1-Turbo-32K | 44 | #288 |
| 7 | GPT-4.1 (2025-04-14) | 42 | #257 |
| 8 | DeepSeek V3 (0324) | 42 | #349 |
| 9 | GLM-4.5 | 42 | #283 |
| 10 | Kimi K2 | 40 | #252 |
| 11 | MiniMax M1 80k | 24 | #437 |
| 12 | GLM 4.5 Air | 22 | #403 |
| 13 | MiniMax-Text-01 | 22 | #519 |
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-hard-multi-turn-compassbench-2507 · How It Works · Data refreshed daily, snapshot 2026-11-05.