COLLIE: leaderboard
2,080 constrained text-generation instances across 13 constraint structures at word, sentence, paragraph and passage level, from Princeton NLP (2023); outputs are checked programmatically.
Metric: Score (%). Source: arxiv.org. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 Mini (High) | 98.7 | #272 (O3 Mini) |
| 2 | O1 (High) | 95.3 | #183 (O1) |
| 3 | GPT-4.5 | 72.3 | #274 |
| 4 | GPT-4.1 | 65.8 | #248 |
| 5 | GPT-4.1 Mini | 54.6 | #355 |
| 6 | GPT-4o Mini | 52.7 | #599 |
| 7 | GPT-4o (2024-11-20) | 50.2 | #381 |
| 8 | GPT-4.1 Nano | 42.5 | #731 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=collie · How It Works · Data refreshed daily, snapshot 2026-10-20.