t2-bench: leaderboard
Agentic tool-use benchmark for selecting and sequencing tools under task constraints, with emphasis on reliable execution rather than single-call function use.
Metric: Score (%). Source: arxiv.org. Source: llm-stats.com. Saturation forecast: Estimated already saturated. 23 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 90.2 | #99 |
| 2 | GLM-5 | 89.7 | #141 |
| 3 | Qwen 3.5 397B A17B | 86.7 | #146 |
| 4 | Gemma 4 31B | 86.4 | #200 |
| 5 | Gemma 4 26B A4B | 85.5 | #358 |
| 6 | Gemini 3 Pro | 85.4 | #80 |
| 7 | Qwen 3.5 35B A3B | 81.2 | #256 |
| 8 | DeepSeek V3.2 | 80.3 | #205 |
| 9 | DeepSeek V3.2 Speciale | 80.3 | #128 |
| 10 | DeepSeek V3.2 (Thinking) | 80.2 | #205 (DeepSeek V3.2) |
| 11 | Qwen 3.5 4B | 79.9 | #475 |
| 12 | Qwen 3.5 122B A10B | 79.5 | #173 |
| 13 | Qwen 3.5 9B | 79.1 | #379 |
| 14 | Qwen 3.5 27B | 79 | #187 |
| 15 | Qwen 3 Max | 74.8 | #208 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=t2-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.