TOBench: leaderboard
Tool-using agents are increasingly expected to operate across realistic professional workflows, where they must interpret multimodal inputs, coordinate external tools, inspect intermediate artifacts, and revise their actions before producing a final result. Existing benchmarks, however, often evaluate tool use, computer use, and multimodal reasoning in isolation, leaving a gap between benchmark se
Metric: Avg. (self-reported). Source: benchmarklist.com. Saturation forecast: Around August 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Plus | 41 | #125 |
| 2 | Claude Opus 4.6 | 32 | #63 |
| 3 | Gemini 3 Pro | 32 | #80 |
| 4 | Qwen 3.6 Plus | 32 | #120 |
| 5 | Kimi K2.5 | 31 | #144 |
| 6 | Gemini 3.1 Pro (Preview) | 30 | #58 |
| 7 | Claude Haiku 4.5 | 27 | #280 |
| 8 | GPT-5 | 26.8 | #98 |
| 9 | Claude Sonnet 4.6 | 23 | #90 |
| 10 | GPT-5.2 | 22 | #110 |
| 11 | GLM-4.6V | 14 | #316 |
| 12 | Gemma 4 31B | 9 | #200 |
| 13 | MiniMax-M2.5 | 7 | #303 |
| 14 | GPT-4o | 6.12 | #343 |
| 15 | Qwen 3 VL 8B Instruct | 4 | #420 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tobench · How It Works · Data refreshed daily, snapshot 2026-10-20.