ClosureBench - L2 Chain: leaderboard
Metric: Accuracy (%; lenient scoring of one JSON answer per question; L2 chained queries (two to three steps, 400 instances); held-out split of 1,300 generated graph questions over graph sizes 4-10, temperature 0; direct mode: answer immediately). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O3 | 97.2 |
| 2 | Gemini 2.5 Flash | 89.8 |
| 3 | Claude Sonnet 4 | 87 |
| 4 | GPT-4.1 | 55.8 |
| 5 | GPT-4.1 Mini | 49.5 |
| 6 | Qwen 3.5 9B | 46.2 |
| 7 | Qwen 3 4B | 43 |
| 8 | Qwen 3.5 2B | 31.8 |
| 9 | Qwen 2.5 1.5B Instruct | 22.2 |
Interactive version: theaggregate.ai/benchmark?slug=closurebench-l2-chain · How It Works · Data refreshed daily, snapshot 2026-09-26.