Time Puzzles - Six-condition Exact Match: leaderboard
Exact-match temporal puzzle accuracy over 600 generated puzzles averaged across six solution-count conditions, original paper v3 Table 1. Table 4 supplies dated OpenAI snapshots and exact open-weight checkpoints; reasoning variants remain separate. This series reports Exact-match Accuracy (%). Source: https://arxiv.org/html/2601.07148v3.
Metric: Exact-match Accuracy (%). Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 55.33 | #105 |
| 2 | DeepSeek V3.2 (Thinking) | 36.67 | #210 (DeepSeek V3.2) |
| 3 | GPT-4.1 | 35.17 | #258 |
| 4 | DeepSeek V3.2 (Non-reasoning) | 30.5 | #210 (DeepSeek V3.2) |
| 5 | GPT-5 Nano | 15.83 | #439 |
| 6 | Qwen 3 14B | 13.17 | #542 |
| 7 | Qwen 3 8B | 10.83 | #678 |
| 8 | GPT-OSS-20B | 10.83 | #523 |
| 9 | GPT-4.1 Nano | 9.33 | #730 |
| 10 | Qwen 3 4B 2507 (Thinking) | 5.67 | #527 |
| 11 | Ministral-3-8B-Reasoning-2512 | 4.27 | #737 |
| 12 | Ministral-3-8B-Instruct-2512 | 3.5 | #663 |
| 13 | Qwen 3 4B 2507 Instruct | 3 | #768 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=time-puzzles-six-condition-exact-match · How It Works · Data refreshed daily, snapshot 2026-11-05.