WildToolBench: leaderboard
Metric: Session accuracy (%): share of the 256 dialogues in which all four tasks are completed correctly, on WildToolBench (256 multi-turn scenarios, four user tasks each, 1,024 tasks, built from real user-log patterns with expert-annotated tool calls), each model through its native function-call format with default decoding; higher is better. Source: arxiv.org. Saturation forecast: Around 2032. 57 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.0 Flash (Thinking) | 14.45 |
| 2 | Gemini 2.5 Pro | 14.06 |
| 3 | Claude Sonnet 4 | 12.5 |
| 4 | O1 | 12.11 |
| 5 | GLM-4.5 | 12.11 |
| 6 | GPT-4o | 11.72 |
| 7 | Claude 3.7 Sonnet | 11.33 |
| 8 | Kimi K2 | 10.55 |
| 9 | Grok 4 | 10.16 |
| 10 | O3 | 10.16 |
| 11 | Qwen 3 30B A3B (Non-reasoning) | 9.77 |
| 12 | DeepSeek R1 | 9.38 |
| 13 | DeepSeek V3 | 9.38 |
| 14 | Qwen 3 14B (Reasoning) | 9.38 |
| 15 | Claude Opus 4.1 | 9.38 |
Interactive version: theaggregate.ai/benchmark?slug=wildtoolbench · How It Works · Data refreshed daily, snapshot 2026-10-07.