O3 (2025-04-16): benchmark results
April 16, 2025 O3 snapshot, tracked when sources report the dated API model. Provider: OpenAI. Released 2025-04-16. Access: API.
Aggregate Rating 1748 ± 7, rank #121 of 1656 rated models, from 1068 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AEPC-QA (Closed-Book) | 76.13 | Accuracy (%) answering from the model's own knowledge on AEP | 100 |
| AEPC-QA (RAG) | 78.68 | Accuracy (%) with the authors' retrieval-augmented pipeline | 100 |
| Arena-Hard v2 | 85.9 | Win Rate (%) | 100 |
| Arena-Hard v2 (GPT-4.1 Judge) | 87 | Win Rate (%) | 100 |
| EuroEval French NLU - Allocine | 98.19 | Matthews correlation coefficient (%) | 100 |
| FactArena | 1320.14 | Elo rating from pairwise judged comparisons | 100 |
| FactArena - Claim Extraction | 1188.11 | Elo rating from pairwise judged comparisons | 100 |
| FactArena - Evidence Retrieval | 1351.95 | Elo rating from pairwise judged comparisons | 100 |
| FactArena - Justification Helpfulness | 1294.67 | Elo rating from pairwise judged comparisons | 100 |
| FactArena - Justification Informativeness | 1331.77 | Elo rating from pairwise judged comparisons | 100 |
| FactArena - Justification Readability | 1268.83 | Elo rating from pairwise judged comparisons | 100 |
| FactArena - Justification Soundness | 1292.55 | Elo rating from pairwise judged comparisons | 100 |
Interactive version: theaggregate.ai/model?slug=o3-2025-04-16 · How It Works · Data refreshed daily, snapshot 2026-10-20.