AgentEscapeBench - Difficulty 20: leaderboard
Metric: Success rate (%; share of the 60 difficulty-20 instances (dependency DAGs of 20 tool and item nodes) whose deterministic final flag the agent submits; single run per instance, official APIs at temperature 0.7, multi-turn tool calling with incremental disclosure of hidden nodes). Source: arxiv.org. Saturation forecast: Around December 2026. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 71 |
| 2 | Gemini 3.1 Pro (Preview) | 60 |
| 3 | GPT-5.4 | 53.3 |
| 4 | Kimi K2.5 | 31.7 |
| 5 | GPT-5 | 20 |
| 6 | Gemini 3 Flash (Preview) | 18.3 |
| 7 | Seed 2.0 Pro | 15 |
| 8 | MiniMax-M2 | 5 |
Interactive version: theaggregate.ai/benchmark?slug=agentescapebench-difficulty-20 · How It Works · Data refreshed daily, snapshot 2026-09-26.