ClawArena-Team - Tool-Permission Precision: leaderboard
Metric: Tool-permission precision (%): per subagent, the share of granted tool types actually used, averaged; 41 multi-turn multimodal scenarios with 258 evaluation rounds; a text-only main agent with partial workspace access creates, empowers and schedules a fixed locally served pool of LLM, VLM and omni subagents; execution-based checks, no LLM judge; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 79.9 |
| 2 | Gemma 4 31B | 79 |
| 3 | Claude Sonnet 4.6 | 77.3 |
| 4 | GPT-5.4 | 77.3 |
| 5 | Kimi K2.6 | 76.8 |
| 6 | Gemini 3.1 Pro (Preview) | 76.4 |
| 7 | Claude Fable 5 | 76.4 |
| 8 | DeepSeek V4 Pro | 73.9 |
| 9 | Qwen 3.6 27B | 72.5 |
| 10 | Gemini 3.5 Flash | 69.8 |
| 11 | GLM-5.2 | 67.4 |
| 12 | GLM-4.7 Flash | 22.4 |
Interactive version: theaggregate.ai/benchmark?slug=clawarena-team-tool-permission-precision · How It Works · Data refreshed daily, snapshot 2026-09-29.