CAPA (No History) - First-Turn Success: leaderboard
Metric: First-turn executable success (%; share of the 300 held-out coding sessions whose first assistant turn is an accepted submission, with no clarification asked; no history: the model sees only the new ambiguous request). Source: arxiv.org. Saturation forecast: Around 2040. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.8 | 24.3 |
| 2 | Kimi K2.6 | 22.3 |
| 3 | GLM-5.2 | 18.7 |
| 4 | Qwen 3 8B | 18.3 |
| 5 | Qwen 3.7 Max | 17.7 |
| 6 | Gemini 3.5 Flash | 17 |
| 7 | DeepSeek V4 Pro | 15.3 |
| 8 | Qwen 3.5 27B | 10.7 |
| 9 | Claude Sonnet 4.6 | 6.3 |
| 10 | Llama 3.3 70B Instruct | 5.7 |
| 11 | GPT-5.6 Sol | 2.7 |
| 12 | GPT-5.5 | 2.3 |
Interactive version: theaggregate.ai/benchmark?slug=capa-no-history-first-turn-success · How It Works · Data refreshed daily, snapshot 2026-09-29.