Model comparison
o3-mini vs o4-mini
o4-mini is the stronger model overall, scoring 41.6 to 36.7 on the Noometry Index.
Last verified . 42 shared benchmarks.
Summary
- They share 42 benchmarks with published results for both. o3-mini scores higher in 0 categories and o4-mini in 9 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where o4-mini leads 40.8 to 28.1.
- The biggest single-benchmark swing is Fiction.LiveBench: 50% for o3-mini and 77.8% for o4-mini.
- Both cost about the same: $1.10 input and $4.40 output per million tokens.
Side by side
| o3-mini | o4-mini | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 36.7 | 41.6 |
| Released | 2024-12-20 | 2025-04-16 |
| Weights | Proprietary | Proprietary |
| Context window | 200K | 200K |
| Max output | 100K | 100K |
| Input $ / M tokens | $1.10 | $1.10 |
| Output $ / M tokens | $4.40 | $4.40 |
| Results tracked | 51 | 60 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
o3-mini: 40.8 (#132), o4-mini: 40.9 (#127)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| Aider Polyglot | 60.4% | 72% |
| GSO | 1.3% | 3.6% |
| WeirdML | 43.7% | 52.6% |
| LMArena Coding | 1378 | 1368 |
| CadEval | 54% | 62% |
| SWE-bench Verified (bash only) | — | 45% |
| SciCode | 39.8% | — |
| LiveBench Coding | 82.7% | — |
| ALE-Bench | — | 826.17 |
| AlgoTune | — | 1.72 |
Agentic & Tool Use o4-mini leads
o3-mini: 29.6 (#84), o4-mini: 32.6 (#61)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 53.2% |
| GDPval | — | 25.3% |
| Cybench | 22.5% | — |
| METR Time Horizons | — | 63.9% |
Reasoning o4-mini leads
o3-mini: 16.3 (#305), o4-mini: 24.6 (#162)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| ARC-AGI-2 | 3% | 6.1% |
| SimpleBench | 22.8% | 38.7% |
| ARC-AGI-1 | 34.5% | 58.7% |
| CritPt | 0.3% | 0.6% |
| Chess Puzzles | 17% | 26% |
| LMArena Hard Prompts | 1366 | 1351 |
| Mystery Game Puzzles | 7% | 5% |
| DTBench | 68.8% | 77.6% |
| LMCA | 19% | 26.5% |
| Epoch Capabilities Index | 140.34 | 145.64 |
| ForecastBench | 59.6 | 61.8 |
| Kagi LLM Benchmark | — | 67.6% |
| EnigmaEval | — | 9.2% |
| LiveBench Reasoning | 89.6% | — |
| LiveBench Data Analysis | 70.6% | — |
| LiveBench | 75.9% | — |
Math o4-mini leads
o3-mini: 28.1 (#244), o4-mini: 40.8 (#89)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| FrontierMath (Tiers 1-3) | 18.6% | 36.1% |
| FrontierMath Tier 4 | 0% | 4.9% |
| OTIS Mock AIME 2024-2025 | 76.9% | 81.7% |
| LMArena Math | 1396 | 1389 |
| MATH Level 5 | 96.5% | 97.8% |
| FrontierMath (Feb 2025 set) | 12.4% | 24.8% |
| FrontierMath Tier 4 (v1) | 4.2% | 6.3% |
| Omni-MATH | — | 72% |
| LiveBench Math | 77.3% | — |
Knowledge o4-mini leads
o3-mini: 38.3 (#146), o4-mini: 43.6 (#91)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| GPQA Diamond | 77% | 79.6% |
| SimpleQA Verified | 15.3% | 19.6% |
| Confabulations | 17.9% | 15.8% |
| LMArena Expert | 1364 | 1343 |
| Humanity's Last Exam | — | 18.1% |
| MMLU-Pro | — | 82% |
| Vectara Hallucination Rate | — | 18.6% |
| GPQA (HELM) | — | 73.5% |
Multimodal Not comparable
o3-mini: —, o4-mini: 40.2 (#49)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| LMArena Vision | — | 1194 |
| GeoBench | — | 64% |
| VPCT | — | 57.5% |
Multilingual o4-mini leads
o3-mini: 45.7 (#164), o4-mini: 47.0 (#154)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| LMArena Non-English | 1319 | 1337 |
| LMArena Chinese | 1379 | 1354 |
| LMArena French | 1334 | 1364 |
| LMArena German | 1303 | 1336 |
| LMArena Japanese | 1286 | 1308 |
| LMArena Korean | 1314 | 1312 |
| LMArena Russian | 1304 | 1334 |
| LMArena Spanish | 1321 | 1347 |
Instruction Following Too close to call
o3-mini: 75.1 (#72), o4-mini: 75.2 (#68)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| LMArena Instruction Following | 1337 | 1321 |
| LiveBench Instruction Following | 84.4% | — |
| IFEval | — | 92.8% |
Long Context o4-mini leads
o3-mini: 33.8 (#256), o4-mini: 45.5 (#33)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| Fiction.LiveBench | 50% | 77.8% |
| LMArena Longer Query | 1343 | 1315 |
Writing & Preference o4-mini leads
o3-mini: 50.3 (#182), o4-mini: 54.0 (#152)
| Benchmark | o3-mini | o4-mini |
|---|---|---|
| LMArena Text | 1337 | 1353 |
| LMArena Creative Writing | 1286 | 1294 |
| Short-Story Creative Writing | 61.7% | 75% |
| LMArena Multi-Turn | 1320 | 1350 |
| WildBench | — | 85.4% |
| LiveBench Language | 50.7% | — |
Frequently asked questions
Is o3-mini better than o4-mini?
o4-mini is the stronger model overall, scoring 41.6 to 36.7 on the Noometry Index.
Which is cheaper, o3-mini or o4-mini?
o4-mini is cheaper. It lists at $1.10 per million input tokens and $4.40 per million output tokens; o3-mini lists at $1.10 and $4.40.
Is o3-mini or o4-mini better for coding?
They score almost the same on coding (40.8 vs 40.9); test both on your own repository before choosing.
Which has the bigger context window?
Both accept 200K tokens.
How many benchmarks do o3-mini and o4-mini share?
42 benchmarks have published results for both models. o3-mini has 51 scored results on Noometry and o4-mini has 60.