CCTU (Thinking): leaderboard
Metric: Perfect solve rate (%): the share of cases in which the model solves every subquery without ever violating a constraint, on the 200 CCTU test cases (FTRL tool-use tasks rewritten with on average seven constraints from 12 categories over resources, behavior, toolsets and responses; locally executable tools), an executable validator checking every step and returning violation feedback for revision, thinking mode enabled, default API settings otherwise, mean of 3 runs; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 (Thinking) | 18.17 | #105 (GPT-5.2) |
| 2 | GPT-5.1 (Thinking) | 16.5 | #131 (GPT-5.1) |
| 3 | Claude Opus 4.6 (Thinking) | 14.83 | #60 (Claude Opus 4.6) |
| 4 | Seed 2.0 Pro (Thinking) | 14.83 | #111 (Seed 2.0 Pro) |
| 5 | Qwen 3.5 Plus (Thinking) | 10.67 | #123 (Qwen 3.5 Plus) |
| 6 | Gemini 3 Pro | 10.5 | #77 |
| 7 | DeepSeek V3.2 (Thinking) | 9 | #198 (DeepSeek V3.2) |
| 8 | O3 | 8.17 | #121 |
| 9 | Kimi K2.5 (Thinking) | 7.67 | #139 (Kimi K2.5) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=cctu-thinking · How It Works · Data refreshed daily, snapshot 2026-10-11.