ToolMaze (Failure-Aware Prompt): leaderboard

Metric: Composite score (%): mean of task success over the clean and four perturbed settings, perturbation recovery rate, and one minus recovery cost, each macro-averaged over the four topology levels; 400 DAG-structured tool-use tasks (four topology levels) over 270 tools, each also run under four injected tool-failure modes (explicit or implicit, transient or permanent); sandboxed, 25 agent steps, temperature 1, a failure-aware prompt that warns of tool failures and outlines recovery strategies (the clean runs keep the standard prompt); higher is better. Source: arxiv.org. Saturation forecast: Around May 2027. 9 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)52.95
2GPT-5.549.96
3Claude Sonnet 4.646.14
4DeepSeek V4 Pro45.09
5GLM-5.138.14
6Qwen 3.5 397B A17B35.56
7Qwen 3.6 27B34.13
8Qwen 3.5 35B A3B32.56
9MiniMax-M2.727.5

Interactive version: theaggregate.ai/benchmark?slug=toolmaze-failure-aware-prompt · How It Works · Data refreshed daily, snapshot 2026-09-29.