ToolRobustBench (Overall): leaderboard
Metric: Success rate (0-1, native; mean tool-call success across the four perturbation families and three severities; over ToolRobustBench single-family perturbation instances (clean, light, medium, heavy); binary per-instance tool-call success). Source: arxiv.org. Saturation forecast: Around December 2026. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 0.77 |
| 2 | DeepSeek V4 Pro | 0.75 |
| 3 | GPT-5.4 Mini | 0.71 |
| 4 | Qwen Plus | 0.71 |
| 5 | Gemini 2.5 Pro | 0.69 |
| 6 | DeepSeek V4 Flash | 0.67 |
| 7 | Gemini 2.5 Flash | 0.66 |
Interactive version: theaggregate.ai/benchmark?slug=toolrobustbench-overall · How It Works · Data refreshed daily, snapshot 2026-09-26.