Model comparison
o3 vs Phi-4
o3 is the stronger model overall, scoring 47.5 to 31.2 on the Noometry Index. Phi-4 costs 40× less per token, which makes it the better buy when o3's lead doesn't matter for your workload.
Last verified . 25 shared benchmarks.
MicrosoftSummary
- They share 25 benchmarks with published results for both. o3 scores higher in 9 categories and Phi-4 in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where o3 leads 50.2 to 20.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 84.4% for o3 and 13.8% for Phi-4.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $2 / $8 for o3.
- o3 accepts more context: 200K tokens versus 128K.
- Phi-4 has downloadable open weights; the other is API-only.
Side by side
| o3 | Phi-4 | |
|---|---|---|
| Provider | OpenAI | Microsoft |
| Noometry Index | 47.5 | 31.2 |
| Released | 2025-04-16 | 2024-12-11 |
| Weights | Proprietary | Open |
| Context window | 200K | 128K |
| Max output | 100K | 4K |
| Input $ / M tokens | $2 | $0.07 |
| Output $ / M tokens | $8 | $0.14 |
| Results tracked | 63 | 37 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding o3 leads
o3: 46.8 (#64), Phi-4: 34.4 (#239)
| Benchmark | o3 | Phi-4 |
|---|---|---|
| LMArena Coding | 1408 | 1231 |
| SWE-bench Verified | 62.3% | — |
| SWE-bench Verified (bash only) | 58.4% | — |
| Aider Polyglot | 81.3% | — |
| GSO | 8.8% | — |
| WeirdML | 52.4% | — |
| BigCodeBench Instruct | — | 45.5% |
| LiveBench Coding | — | 30.7% |
| BigCodeBench Complete | — | 55.4% |
| CadEval | 74% | — |
| ALE-Bench | 933.55 | — |
Agentic & Tool Use o3 leads
o3: 34.5 (#44), Phi-4: 22.8 (#128)
| Benchmark | o3 | Phi-4 |
|---|---|---|
| Berkeley Function Calling Leaderboard | 63% | 28.8% |
| GDPval | 30.8% | — |
| DeepResearch Bench | 45.2% | — |
| OSWorld | 23% | — |
| BALROG | — | 11.6% |
| LMArena Search | 1144 | — |
| METR Time Horizons | 65.4% | — |
Reasoning o3 leads
o3: 32.0 (#78), Phi-4: 17.7 (#291)
| Benchmark | o3 | Phi-4 |
|---|---|---|
| Chess Puzzles | 38% | 1% |
| LMArena Hard Prompts | 1402 | 1220 |
| Epoch Capabilities Index | 146.86 | 130.42 |
| ARC-AGI-2 | 6.5% | — |
| SimpleBench | 53.1% | — |
| Kagi LLM Benchmark | 67.6% | — |
| ARC-AGI-1 | 60.8% | — |
| CritPt | 1.4% | — |
| EnigmaEval | 13.1% | — |
| LiveBench Reasoning | — | 47.8% |
| Mystery Game Puzzles | 29% | — |
| DTBench | 84.8% | — |
| LiveBench Data Analysis | — | 45.2% |
| LMCA | 39.7% | — |
| ForecastBench | 62.5 | — |
| LiveBench | — | 41.6% |
Math o3 leads
o3: 50.2 (#58), Phi-4: 20.8 (#285)
| Benchmark | o3 | Phi-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 84.4% | 13.8% |
| LMArena Math | 1426 | 1246 |
| MATH Level 5 | 97.8% | 64.9% |
| FrontierMath (Tiers 1-3) | 33.3% | — |
| Omni-MATH | 71.4% | — |
| LiveBench Math | — | 42% |
| FrontierMath (Feb 2025 set) | 18.7% | — |
| FrontierMath Tier 4 (v1) | 2.1% | — |
Knowledge o3 leads
o3: 54.6 (#52), Phi-4: 32.6 (#209)
| Benchmark | o3 | Phi-4 |
|---|---|---|
| GPQA Diamond | 81.8% | 56.1% |
| Confabulations | 14.4% | 29.4% |
| LMArena Expert | 1402 | 1203 |
| Humanity's Last Exam | 20.3% | — |
| SimpleQA Verified | 49.4% | — |
| MMLU-Pro | 85.9% | — |
| Vectara Hallucination Rate | — | 3.7% |
| GPQA (HELM) | 75.3% | — |
| MMLU | — | 84.8% |
Multimodal Not comparable
o3: 41.4 (#36), Phi-4: —
| Benchmark | o3 | Phi-4 |
|---|---|---|
| LMArena Vision | 1214 | — |
| GeoBench | 74% | — |
| VPCT | 52% | — |
Multilingual o3 leads
o3: 51.7 (#105), Phi-4: 37.2 (#237)
| Benchmark | o3 | Phi-4 |
|---|---|---|
| LMArena Non-English | 1401 | 1197 |
| LMArena Chinese | 1437 | 1212 |
| LMArena French | 1430 | 1224 |
| LMArena German | 1420 | 1222 |
| LMArena Japanese | 1403 | 1158 |
| LMArena Korean | 1370 | 1151 |
| LMArena Russian | 1406 | 1209 |
| LMArena Spanish | 1395 | 1234 |
Instruction Following o3 leads
o3: 72.8 (#127), Phi-4: 60.4 (#251)
| Benchmark | o3 | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1368 | 1201 |
| LiveBench Instruction Following | — | 58.4% |
| IFEval | 86.9% | — |
Long Context o3 leads
o3: 53.3 (#6), Phi-4: 36.9 (#226)
| Benchmark | o3 | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1372 | 1217 |
| Fiction.LiveBench | 88.9% | — |
| CL-bench | 17.8% | — |
Writing & Preference o3 leads
o3: 63.5 (#64), Phi-4: 40.5 (#244)
| Benchmark | o3 | Phi-4 |
|---|---|---|
| LMArena Text | 1410 | 1217 |
| LMArena Creative Writing | 1359 | 1182 |
| Short-Story Creative Writing | 83.9% | 62.6% |
| LMArena Multi-Turn | 1405 | 1206 |
| EQ-Bench Creative Writing | 1676 | — |
| WildBench | 86.1% | — |
| LiveBench Language | — | 25.6% |
Frequently asked questions
Is o3 better than Phi-4?
o3 is the stronger model overall, scoring 47.5 to 31.2 on the Noometry Index. Phi-4 costs 40× less per token, which makes it the better buy when o3's lead doesn't matter for your workload.
Which is cheaper, o3 or Phi-4?
Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; o3 lists at $2 and $8.
Is o3 or Phi-4 better for coding?
o3 scores higher on coding benchmarks: 46.8 versus 34.4 in the Noometry coding category.
Which has the bigger context window?
o3 does, with 200K tokens against 128K.
How many benchmarks do o3 and Phi-4 share?
25 benchmarks have published results for both models. o3 has 63 scored results on Noometry and Phi-4 has 37.