O1: benchmark results
OpenAI's first-generation o-series reasoning model for difficult multi-step tasks. Provider: OpenAI. Released 2024-12-05. Access: API.
Aggregate Rating 1705 ± 5, rank #183 of 1656 rated models, from 638 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AbstentionBench - false premise - CoCoNot/False presumptions - F1 | 89.57 | Abstention F1 (%) | 100 |
| AbstentionBench - false premise - CoCoNot/False presumptions - Recall | 87.95 | Abstention Recall (%) | 100 |
| AbstentionBench - false premise - QAQA - F1 | 74.13 | Abstention F1 (%) | 100 |
| AbstentionBench - false premise - QAQA - Recall | 63.86 | Abstention Recall (%) | 100 |
| AbstentionBench - underspecified context - GPQA-Diamond - F1 | 87.32 | Abstention F1 (%) | 100 |
| AbstentionBench - underspecified context - GSM8K - F1 | 97.73 | Abstention F1 (%) | 100 |
| AbstentionBench - underspecified context - GSM8K - Recall | 95.63 | Abstention Recall (%) | 100 |
| AbstentionBench - underspecified context - UMWP - F1 | 86.38 | Abstention F1 (%) | 100 |
| AbstentionBench - underspecified context - UMWP - Recall | 76.11 | Abstention Recall (%) | 100 |
| AbstentionBench - underspecified context - WorldSense - Recall | 98.96 | Abstention Recall (%) | 100 |
| AidanBench | 6054 | Novel Answers | 100 |
| BlueBench - Knowledge | 71.43 | Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=o1 · How It Works · Data refreshed daily, snapshot 2026-10-20.