Up to four models
o4-mini vs o3-mini
On the index (GPT-4 = 100), o4-mini scores 141 and o3-mini 131. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
o4-mini leads o3-mini by 10 points, ahead on 3 of 3 shared core tests.
Test by testthird-party results
o4-minio3-mini
Every result as a table
| Test | o4-mini | o3-mini |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 79.6% | 77.0% |
| GPQA Diamond (Vals AI) | 74.5% | 75.5% |
| MMLU Pro | 80.6% | 78.7% |
| Fiction.LiveBench | 77.8% | 50.0% |
| SimpleBench | 38.7% | 22.8% |
| SimpleQA Verified | 19.6% | 15.3% |
| Maths | ||
| FrontierMath (tiers 1-3) | 36.1% | 18.6% |
| AIME-style maths (OTIS mock) | 81.7% | 76.9% |
| MATH (level 5) | 97.8% | 96.5% |
| MATH 500 | 94.2% | 91.8% |
| MGSM | 93.4% | 91.3% |
| Chess puzzles | 22.0% | 15.0% |
| FrontierMath (2025 set) | 24.8% | 12.4% |
| FrontierMath tier 4 (2025 set) | 6.3% | 4.2% |
| FrontierMath (tier 4) | 4.9% | 0.0% |
| Coding | ||
| Aider Polyglot | 72.0% | 60.4% |
| LiveCodeBench | 82.2% | 71.5% |
| CadEval | 62.0% | 54.0% |
| GSO-Bench | 3.6% | 1.3% |
| WeirdML | 52.6% | 43.8% |
| Agents | ||
| DTBench | 77.6% | 68.8% |
| LMCA | 26.5% | 19.0% |
| Mystery game puzzles | 5.0% | 7.0% |
| Professional work | ||
| CorpFin v2 | 59.0% | 45.3% |
| LegalBench | 79.2% | 71.5% |
| TaxEval v2 | 74.8% | 69.4% |
| Novel problems | ||
| ARC-AGI-2 | 6.1% | 3.0% |
| ARC-AGI-1 | 58.7% | 34.5% |
Task length · METR38 sec
Tests at least two of these models have taken. Bold rows are the core tests.