OpenClawProBench: leaderboard
Agentic intelligence benchmark with 102 active scenarios testing planning, tool use, constraints, error recovery, synthesis, and safety. Deterministic grading across 3 trials per model.
Metric: Overall Score (%). Source: suyoumo.github.io. Status: years away from saturation. 68 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Kimi K3 | 81.6 |
| 2 | GLM-5.2 | 81.3 |
| 3 | MiniMax-M3 | 75.1 |
| 4 | Step 3.7 Flash | 72 |
| 5 | Ernie 5.1 | 70.7 |
| 6 | Qwen 3.5 397B A17B | 70.4 |
| 7 | Qwen 3.5 Plus | 70.1 |
| 8 | DeepSeek V4 Pro | 69.6 |
| 9 | GPT-5.5 (xHigh) | 69.3 |
| 10 | GLM-5.1 | 69 |
| 11 | MiMo-V2.5-Pro | 68.5 |
| 12 | Seed 2.0 Pro | 68.3 |
| 13 | GPT-5.4 (xHigh) | 68 |
| 14 | DeepSeek V3.2 | 67.6 |
| 15 | DeepSeek V4 Flash | 67.6 |
Interactive version: theaggregate.ai/benchmark?slug=openclawprobench · How It Works · Data refreshed daily, snapshot 2026-09-05.