GPT-4o (Mar 2025): benchmark results
March 2025 GPT-4o snapshot kept separate from other GPT-4o rows because benchmark sources report distinct scores. Provider: OpenAI. Released 2025-03-01. Access: API.
Aggregate Rating 1641 ± 24, rank #330 of 1656 rated models, from 31 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LLM Emergent Collusion | 23 | Collusion Rate (%) | 91.7 |
| Elimination Game (Lechmazur) | 5.5 | TrueSkill μ | 89.8 |
| OpenAI MentalHealthBench - Behavior - User Agency | 59.51 | Axis score, normalized to its possible range (%) | 71.4 |
| AA MMLU-Pro | 80.29 | Accuracy (%) | 68.1 |
| OpenAI MentalHealthBench - Behavior - Communication | 66.84 | Axis score, normalized to its possible range (%) | 64.3 |
| BAZAAR (Lechmazur) | 5.89 | TrueSkill μ | 56 |
| AA LiveCodeBench | 42.54 | Pass@1 (%) | 50.4 |
| Generalization V1 (Lechmazur) | 1.97 | Avg Rank (lower is better) | 43.1 |
| OpenAI MentalHealthBench - Behavior - Interpretation | 56.93 | Axis score, normalized to its possible range (%) | 42.9 |
| NYT Connections Older Models | 11.8 | Score (%) | 42 |
| AA GPQA Diamond | 65.45 | Accuracy (%) | 40.5 |
| Step Game (Lechmazur) | 1.55 | TrueSkill μ | 39.2 |
Interactive version: theaggregate.ai/model?slug=gpt-4o-mar-2025 · How It Works · Data refreshed daily, snapshot 2026-10-20.