O3: benchmark results
OpenAI's o-series o3 reasoning model for difficult multi-step tasks. Provider: OpenAI. Released 2025-04-16. Access: API.
Aggregate Rating 1743 ± 3, rank #124 of 1656 rated models, from 1195 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentNoiseBench (VitaBench, Tool Noise) | 46 | Success on a stratified 25 percent sample of VitaBench tasks | 100 |
| BAZAAR (Lechmazur) | 7.37 | TrueSkill μ | 100 |
| BLINKG - Schema-Distant - Entity Class | 54 | F1 (%; expert-validated mappings for the ontology class of e | 100 |
| BLINKG - Schema-Distant - Ontology Property | 53 | F1 (%; expert-validated mappings for the ontology property e | 100 |
| BLINKG - Schema-Distant - Related Entity Class | 63 | F1 (%; expert-validated mappings for the class of the entity | 100 |
| BesiegeField - Catapult - Iterative Editing | 9.14 | Mean distance (m) | 100 |
| Beyond Continuity | 97.7 | F1 Pivot (TopiOCQA) (self-reported) | 100 |
| CadEval | 74 | Score (imported) | 100 |
| CharXiv-D | 95 | Score (%) | 100 |
| CharXiv-R | 78.6 | Validation chart-visible reasoning accuracy (%) | 100 |
| ClawForge - State Repair | 33.7 | Strict full-pass accuracy (%) on the 92 tasks whose primary | 100 |
| ClosureBench | 92.8 | Accuracy (%; lenient scoring of one JSON answer per question | 100 |
Interactive version: theaggregate.ai/model?slug=o3 · How It Works · Data refreshed daily, snapshot 2026-10-20.