Model comparison
GPT-4o vs o3
o3 is the stronger model overall, scoring 47.5 to 28.6 on the Noometry Index.
Last verified . 54 shared benchmarks.
Summary
- They share 54 benchmarks with published results for both. GPT-4o scores higher in 0 categories and o3 in 10 categories; 10 gaps are clear of the uncertainty.
- The widest gap is in math, where o3 leads 50.2 to 10.6.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 6.4% for GPT-4o and 84.4% for o3.
- o3 is cheaper at $2 / $8 per million input/output tokens, against $2.50 / $10 for GPT-4o.
- o3 accepts more context: 200K tokens versus 128K.
Side by side
| GPT-4o | o3 | |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 28.6 | 47.5 |
| Released | 2024-05-13 | 2025-04-16 |
| Weights | Proprietary | Proprietary |
| Context window | 128K | 200K |
| Max output | 16K | 100K |
| Input $ / M tokens | $2.50 | $2 |
| Output $ / M tokens | $10 | $8 |
| Results tracked | 72 | 63 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding o3 leads
GPT-4o: 24.8 (#328), o3: 46.8 (#64)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| SWE-bench Verified | 31% | 62.3% |
| SWE-bench Verified (bash only) | 21.6% | 58.4% |
| Aider Polyglot | 45.3% | 81.3% |
| GSO | 0% | 8.8% |
| WeirdML | 25.1% | 52.4% |
| LMArena Coding | 1297 | 1408 |
| CadEval | 26% | 74% |
| BigCodeBench Instruct | 51.1% | — |
| LiveBench Coding | 51.4% | — |
| BigCodeBench Complete | 61.1% | — |
| ALE-Bench | — | 933.55 |
| HumanEval+ | 87.2% | — |
| MBPP+ | 72.2% | — |
Agentic & Tool Use o3 leads
GPT-4o: 21.0 (#141), o3: 34.5 (#44)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| GDPval | 9.9% | 30.8% |
| LMArena Search | 1006 | 1144 |
| METR Time Horizons | 40.8% | 65.4% |
| Berkeley Function Calling Leaderboard | — | 63% |
| TheAgentCompany | 8.6% | — |
| Cybench | 12.5% | — |
| DeepResearch Bench | — | 45.2% |
| OSWorld | — | 23% |
| BALROG | 32.3% | — |
Reasoning o3 leads
GPT-4o: 9.4 (#343), o3: 32.0 (#78)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| ARC-AGI-2 | 0% | 6.5% |
| SimpleBench | 17.8% | 53.1% |
| ARC-AGI-1 | 4.5% | 60.8% |
| CritPt | 0% | 1.4% |
| Chess Puzzles | 13% | 38% |
| EnigmaEval | 0.8% | 13.1% |
| LMArena Hard Prompts | 1281 | 1402 |
| DTBench | 64.5% | 84.8% |
| LMCA | 16.6% | 39.7% |
| Epoch Capabilities Index | 128.97 | 146.86 |
| ForecastBench | 57.7 | 62.5 |
| Kagi LLM Benchmark | — | 67.6% |
| LiveBench Reasoning | 55.8% | — |
| Mystery Game Puzzles | — | 29% |
| LiveBench Data Analysis | 60.9% | — |
| LiveBench | 55.3% | — |
Math o3 leads
GPT-4o: 10.6 (#312), o3: 50.2 (#58)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| FrontierMath (Tiers 1-3) | 0.4% | 33.3% |
| OTIS Mock AIME 2024-2025 | 6.4% | 84.4% |
| Omni-MATH | 29.3% | 71.4% |
| LMArena Math | 1285 | 1426 |
| MATH Level 5 | 53.3% | 97.8% |
| FrontierMath (Feb 2025 set) | 0.3% | 18.7% |
| LiveBench Math | 49.5% | — |
| FrontierMath Tier 4 (v1) | — | 2.1% |
Knowledge o3 leads
GPT-4o: 28.8 (#242), o3: 54.6 (#52)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| GPQA Diamond | 49.2% | 81.8% |
| Humanity's Last Exam | 2.7% | 20.3% |
| SimpleQA Verified | 26% | 49.4% |
| MMLU-Pro | 71.3% | 85.9% |
| Confabulations | 15.3% | 14.4% |
| GPQA (HELM) | 52% | 75.3% |
| LMArena Expert | 1250 | 1402 |
| Vectara Hallucination Rate | 9.6% | — |
| MMLU | 88.1% | — |
Multimodal o3 leads
GPT-4o: 34.5 (#91), o3: 41.4 (#36)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| LMArena Vision | 1137 | 1214 |
| GeoBench | 71% | 74% |
| VPCT | 40% | 52% |
| Video-MME | 71.9% | — |
| ScienceQA | 88.5% | — |
Multilingual o3 leads
GPT-4o: 43.2 (#186), o3: 51.7 (#105)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| LMArena Non-English | 1283 | 1401 |
| LMArena Chinese | 1277 | 1437 |
| LMArena French | 1304 | 1430 |
| LMArena German | 1282 | 1420 |
| LMArena Japanese | 1257 | 1403 |
| LMArena Korean | 1234 | 1370 |
| LMArena Russian | 1286 | 1406 |
| LMArena Spanish | 1292 | 1395 |
Instruction Following o3 leads
GPT-4o: 66.6 (#207), o3: 72.8 (#127)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| IFEval | 81.7% | 86.9% |
| LMArena Instruction Following | 1278 | 1368 |
| LiveBench Instruction Following | 68.6% | — |
Long Context o3 leads
GPT-4o: 39.4 (#179), o3: 53.3 (#6)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| Fiction.LiveBench | 66.7% | 88.9% |
| LMArena Longer Query | 1289 | 1372 |
| CL-bench | — | 17.8% |
Writing & Preference o3 leads
GPT-4o: 52.6 (#166), o3: 63.5 (#64)
| Benchmark | GPT-4o | o3 |
|---|---|---|
| LMArena Text | 1300 | 1410 |
| LMArena Creative Writing | 1292 | 1359 |
| Short-Story Creative Writing | 81.8% | 83.9% |
| WildBench | 82.8% | 86.1% |
| LMArena Multi-Turn | 1302 | 1405 |
| EQ-Bench Creative Writing | — | 1676 |
| LiveBench Language | 47.6% | — |
Frequently asked questions
Is GPT-4o better than o3?
o3 is the stronger model overall, scoring 47.5 to 28.6 on the Noometry Index.
Which is cheaper, GPT-4o or o3?
o3 is cheaper. It lists at $2 per million input tokens and $8 per million output tokens; GPT-4o lists at $2.50 and $10.
Is GPT-4o or o3 better for coding?
o3 scores higher on coding benchmarks: 46.8 versus 24.8 in the Noometry coding category.
Which has the bigger context window?
o3 does, with 200K tokens against 128K.
How many benchmarks do GPT-4o and o3 share?
54 benchmarks have published results for both models. GPT-4o has 72 scored results on Noometry and o3 has 63.