Model comparison
o1 vs Phi-4
o1 is the stronger model overall, scoring 40.6 to 30.9 on the Noometry Index. Phi-4 costs 300× less per token, which makes it the better buy when o1's lead doesn't matter for your workload.
Last verified . 31 shared benchmarks.
MicrosoftSummary
- They share 31 benchmarks with published results for both. o1 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 o1 leads 34.9 to 19.6.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 73.3% for o1 and 13.8% for Phi-4.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $15 / $60 for o1.
- o1 accepts more context: 200K tokens versus 128K.
- Phi-4 has downloadable open weights; the other is API-only.
Side by side
| o1 | Phi-4 | |
|---|---|---|
| Provider | OpenAI | Microsoft |
| Noometry Index | 40.6 | 30.9 |
| Released | 2024-09-12 | 2024-12-11 |
| Weights | Proprietary | Open |
| Context window | 200K | 128K |
| Max output | 100K | 4K |
| Input $ / M tokens | $15 | $0.07 |
| Output $ / M tokens | $60 | $0.14 |
| Results tracked | 52 | 37 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding o1 leads
o1: 45.7 (#64), Phi-4: 33.8 (#237)
| Benchmark | o1 | Phi-4 |
|---|---|---|
| LiveBench Coding | 69.7% | 30.7% |
| LMArena Coding | 1367 | 1231 |
| Aider Polyglot | 61.7% | — |
| WeirdML | 47.6% | — |
| BigCodeBench Instruct | — | 45.5% |
| BigCodeBench Complete | — | 55.4% |
| CadEval | 56% | — |
| HumanEval+ | 89% | — |
| MBPP+ | 80.2% | — |
Agentic & Tool Use o1 leads
o1: 24.3 (#111), Phi-4: 22.6 (#120)
| Benchmark | o1 | Phi-4 |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 28.8% |
| Cybench | 10% | — |
| BALROG | — | 11.6% |
| METR Time Horizons | 51.1% | — |
Reasoning o1 leads
o1: 27.5 (#103), Phi-4: 17.0 (#287)
| Benchmark | o1 | Phi-4 |
|---|---|---|
| Chess Puzzles | 15% | 1% |
| LiveBench Reasoning | 91.6% | 47.8% |
| LMArena Hard Prompts | 1372 | 1220 |
| LiveBench Data Analysis | 65.5% | 45.2% |
| Epoch Capabilities Index | 141.92 | 130.43 |
| LiveBench | 75.7% | 41.6% |
| SimpleBench | 41.7% | — |
| ARC-AGI-1 | 30.7% | — |
| EnigmaEval | 5.7% | — |
| DTBench | 74.7% | — |
| LMCA | 22.3% | — |
Math o1 leads
o1: 34.9 (#150), Phi-4: 19.6 (#279)
| Benchmark | o1 | Phi-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 73.3% | 13.8% |
| LiveBench Math | 80.3% | 42% |
| LMArena Math | 1388 | 1246 |
| MATH Level 5 | 94.7% | 64.9% |
| FrontierMath (Tiers 1-3) | 14.7% | — |
| FrontierMath (Feb 2025 set) | 9.3% | — |
Knowledge o1 leads
o1: 41.5 (#101), Phi-4: 32.5 (#204)
| Benchmark | o1 | Phi-4 |
|---|---|---|
| GPQA Diamond | 76.8% | 56.1% |
| Confabulations | 11.7% | 29.4% |
| LMArena Expert | 1362 | 1203 |
| Humanity's Last Exam | 8% | — |
| SimpleQA Verified | 41.1% | — |
| Vectara Hallucination Rate | — | 3.7% |
| MMLU | — | 84.8% |
Multimodal Not comparable
o1: 33.6 (#87), Phi-4: —
| Benchmark | o1 | Phi-4 |
|---|---|---|
| LMArena Vision | 1168 | — |
| GeoBench | 80% | — |
| VPCT | 37% | — |
| SpatialViz-Bench | 41.4% | — |
Multilingual o1 leads
o1: 48.8 (#137), Phi-4: 37.4 (#232)
| Benchmark | o1 | Phi-4 |
|---|---|---|
| LMArena Non-English | 1359 | 1197 |
| LMArena Chinese | 1394 | 1211 |
| LMArena French | 1343 | 1223 |
| LMArena German | 1337 | 1222 |
| LMArena Japanese | 1346 | 1158 |
| LMArena Korean | 1396 | 1151 |
| LMArena Russian | 1356 | 1209 |
| LMArena Spanish | 1345 | 1234 |
Instruction Following o1 leads
o1: 75.0 (#82), Phi-4: 60.6 (#246)
| Benchmark | o1 | Phi-4 |
|---|---|---|
| LiveBench Instruction Following | 81.5% | 58.4% |
| LMArena Instruction Following | 1368 | 1201 |
Long Context o1 leads
o1: 50.3 (#9), Phi-4: 36.9 (#221)
| Benchmark | o1 | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1378 | 1217 |
| Fiction.LiveBench | 83.3% | — |
Writing & Preference o1 leads
o1: 55.8 (#139), Phi-4: 40.7 (#239)
| Benchmark | o1 | Phi-4 |
|---|---|---|
| LMArena Text | 1366 | 1217 |
| LMArena Creative Writing | 1347 | 1182 |
| Short-Story Creative Writing | 70.2% | 62.6% |
| LMArena Multi-Turn | 1368 | 1206 |
| LiveBench Language | 65.4% | 25.6% |
Frequently asked questions
Is o1 better than Phi-4?
o1 is the stronger model overall, scoring 40.6 to 30.9 on the Noometry Index. Phi-4 costs 300× less per token, which makes it the better buy when o1's lead doesn't matter for your workload.
Which is cheaper, o1 or Phi-4?
Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; o1 lists at $15 and $60.
Is o1 or Phi-4 better for coding?
o1 scores higher on coding benchmarks: 45.7 versus 33.8 in the Noometry coding category.
Which has the bigger context window?
o1 does, with 200K tokens against 128K.
How many benchmarks do o1 and Phi-4 share?
31 benchmarks have published results for both models. o1 has 52 scored results on Noometry and Phi-4 has 37.