General365 - Recursive & Backtracking: leaderboard
Metric: Accuracy (%) on the problems labelled Recursive & Backtracking (a problem may carry several challenge labels), out of 1,460 problems (365 hand-written seeds and 1,095 expansions); numeric answers checked with math-verify, choice and text answers graded by GPT-4.1; temperature 1.0 for reasoning and 0.7 for chat models, highest available reasoning effort; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 26 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash | 69.3 |
| 2 | GLM-4.7 (Thinking) | 68.8 |
| 3 | GLM-5 (Thinking) | 68.2 |
| 4 | Qwen 3.5 397B A17B (Thinking) | 67.9 |
| 5 | Gemini 3 Pro | 66.8 |
| 6 | Qwen 3 Max (Thinking) | 66.8 |
| 7 | DeepSeek V3.2 Speciale | 65.8 |
| 8 | GPT-5 (High) | 65.2 |
| 9 | Kimi K2.5 (Thinking) | 65.2 |
| 10 | DeepSeek V3.2 (Thinking) | 64.9 |
| 11 | GPT-5.1 (High) | 64.4 |
| 12 | GLM-4.6 (Thinking) | 64.1 |
| 13 | O4 Mini (High) | 62.8 |
| 14 | Grok 4.1 Fast (Reasoning) | 62.5 |
| 15 | Kimi K2 (Thinking) | 62.5 |
Interactive version: theaggregate.ai/benchmark?slug=general365-recursive-backtracking · How It Works · Data refreshed daily, snapshot 2026-10-07.