PAIR-Bench - Gap Closure: leaderboard
Gap closure on PAIR-Bench (2026): the percent of the remaining hidden-test pass-rate gap after the first attempt that the best program within the hint budget closes, averaged over initially failing programs. Higher is better.
Metric: Gap closure (%; share of the remaining hidden-test pass-rate gap after the first attempt that the best program within the hint budget closes, averaged over initially failing instances; higher is better). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3.2 | 98.8 | #406 |
| 2 | Gemini 2.5 Flash Lite | 97.51 | #958 |
| 3 | Qwen 3 Coder 30B A3B Instruct | 94.25 | #972 |
| 4 | Ministral 3 14B | 91.73 | #1845 |
| 5 | Llama 3.3 70B Instruct | 89.32 | #1357 |
| 6 | GPT-4o Mini | 89.08 | #1623 |
| 7 | Mistral Small 3.2 | 87.92 | #1591 |
| 8 | Gemma 3 27B | 80.06 | #1631 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pair-bench-gap-closure · How It Works · Data refreshed daily, snapshot 2026-11-13.