Galileo Tool Tasks - tau Long Context: leaderboard
Galileo tau-bench-style long-context tool-use slice, testing whether models keep track of task state and tool arguments over extended interactions.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet (20250219) | 100 | #201 |
| 2 | O1 (2024-12-17) | 98 | #149 |
| 3 | Qwen 2.5 72B Instruct | 92 | #454 |
| 4 | GPT-4o (2024-11-20) | 92 | #381 |
| 5 | Mistral Small 3 | 92 | #627 |
| 6 | Gemini 2.0 Flash (001) | 90 | #360 |
| 7 | Gemini 1.5 Flash | 89 | #618 |
| 8 | O3 Mini (2025-01-31) | 84 | #240 |
| 9 | Mistral Large 2 (Nov) Instruct (2411) | 83 | #465 |
| 10 | Gemini 1.5 Pro | 75 | #445 |
| 11 | Ministral-8B-Instruct-2410 | 73 | #1094 |
| 12 | Llama 3.3 70B Instruct | 72 | #534 |
| 13 | Claude 3.5 Haiku (20241022) | 72 | #588 |
| 14 | Claude 3.5 Sonnet (20241022) | 68 | #291 |
| 15 | GPT-4o Mini | 51 | #599 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=galileo-tool-tasks-tau-long-context · How It Works · Data refreshed daily, snapshot 2026-10-20.