AgentEscapeBench - Difficulty 5: leaderboard
Escape-room style tool use from AgentEscapeBench (Guo et al., 2026): 60 generated puzzles whose hidden dependency graph has 5 tool and item nodes, which an agent must uncover by inspecting entities and calling real functions such as hashing, decoding and archive tools before submitting a verifiable final flag. Success rate in percent, one run per instance at temperature 0.7; higher is better.
Metric: Success rate (%; share of the 60 difficulty-5 instances (dependency DAGs of 5 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: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 96.7 | #121 |
| 2 | Kimi K2.5 | 95 | #252 |
| 3 | Gemini 3.1 Pro (Preview) | 91.7 | #91 |
| 4 | Claude Opus 4.6 | 90 | #100 |
| 5 | Gemini 3 Flash (Preview) | 88.3 | #141 |
| 6 | Qwen 3 235B A22B | 88.3 | #622 |
| 7 | Seed 2.0 Pro | 86.7 | #209 |
| 8 | Qwen 3 Next 80B A3B | 76.7 | #756 |
| 9 | MiniMax-M2 | 70 | #669 |
| 10 | Qwen 3 32B | 60 | #960 |
| 11 | GPT-5 | 56.7 | #163 |
| 12 | Qwen 3 14B | 50 | #1235 |
| 13 | Qwen 3 8B | 28.3 | #1838 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=agentescapebench-difficulty-5 · How It Works · Data refreshed daily, snapshot 2026-11-12.