CirrusBench (Tool-Call) - Pass@2: leaderboard
Metric: Pass@2 success rate (%): share of CirrusBench's 425 multi-turn cloud-customer-service tasks rebuilt from real support logs, the agent replying at each annotated checkpoint and a DeepSeek-V3.2 evaluator checking each reply against the ground-truth response that require tool calls, resolved in at least one of two attempts; higher is better. Source: arxiv.org. Saturation forecast: Around November 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 27.8 | #91 |
| 2 | GPT-5 Mini (2025-08-07) | 22.6 | #165 |
| 3 | Qwen 3 Max | 19.4 | #201 |
| 4 | Qwen 3 235B A22B 2507 (Thinking) | 18.8 | #253 (Qwen 3 235B A22B 2507) |
| 5 | Qwen 3 235B A22B 2507 Instruct | 17.4 | #291 |
| 6 | GPT-5.2 | 16.3 | #105 |
| 7 | DeepSeek V3.2 (Non-reasoning) | 15.5 | #198 (DeepSeek V3.2) |
| 8 | DeepSeek R1 | 10.1 | #245 |
| 9 | GPT-4o (2024-08-06) | 8.7 | #326 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=cirrusbench-tool-call-pass-2 · How It Works · Data refreshed daily, snapshot 2026-10-11.