Agent Planning Benchmark - Tool-Broken Recovery: leaderboard
Metric: Replace rate (%): share of 300 cases in which, after a critical tool returns an error, the model switches to the provided substitute tool (the optimal behaviour, versus altering, retrying or refusing); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) (High) | 78 |
| 2 | Qwen 3 VL 235B A22B Instruct | 77.7 |
| 3 | Claude Sonnet 4.5 | 76.7 |
| 4 | GPT-5 | 76.3 |
| 5 | Gemini 2.5 Pro | 74.7 |
| 6 | Qwen 3 VL 32B Instruct | 68.7 |
| 7 | Qwen 3 VL 30B A3B Instruct | 55.7 |
| 8 | Gemini 2.5 Flash (Non-reasoning) | 44.3 |
| 9 | GPT-4o | 42.3 |
Interactive version: theaggregate.ai/benchmark?slug=agent-planning-benchmark-tool-broken-recovery · How It Works · Data refreshed daily, snapshot 2026-09-29.