Model comparison
GPT-4.1 vs Phi-4
GPT-4.1 is the stronger model overall, scoring 35.9 to 31.2 on the Noometry Index. Phi-4 costs 40× less per token, which makes it the better buy when GPT-4.1's lead doesn't matter for your workload.
Last verified . 24 shared benchmarks.
MicrosoftSummary
- They share 24 benchmarks with published results for both. GPT-4.1 scores higher in 8 categories and Phi-4 in 1 category; 8 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where GPT-4.1 leads 57.6 to 40.5.
- The biggest single-benchmark swing is Berkeley Function Calling Leaderboard: 54% for GPT-4.1 and 28.8% for Phi-4.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $2 / $8 for GPT-4.1.
- GPT-4.1 accepts more context: 1.05M tokens versus 128K.
- Phi-4 has downloadable open weights; the other is API-only.
Side by side
| GPT-4.1 | Phi-4 | |
|---|---|---|
| Provider | OpenAI | Microsoft |
| Noometry Index | 35.9 | 31.2 |
| Released | 2025-04-14 | 2024-12-11 |
| Weights | Proprietary | Open |
| Context window | 1.05M | 128K |
| Max output | 33K | 4K |
| Input $ / M tokens | $2 | $0.07 |
| Output $ / M tokens | $8 | $0.14 |
| Results tracked | 52 | 37 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
GPT-4.1: 34.4 (#238), Phi-4: 34.4 (#239)
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| LMArena Coding | 1391 | 1231 |
| SWE-bench Verified | 48.5% | — |
| SWE-bench Verified (bash only) | 39.6% | — |
| Aider Polyglot | 52.4% | — |
| WeirdML | 39% | — |
| BigCodeBench Instruct | — | 45.5% |
| LiveBench Coding | — | 30.7% |
| BigCodeBench Complete | — | 55.4% |
| CadEval | 42% | — |
| ALE-Bench | 558.1 | — |
Agentic & Tool Use GPT-4.1 leads
GPT-4.1: 34.7 (#43), Phi-4: 22.8 (#128)
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| Berkeley Function Calling Leaderboard | 54% | 28.8% |
| BALROG | — | 11.6% |
Reasoning Phi-4 leads
GPT-4.1: 11.7 (#339), Phi-4: 17.7 (#291)
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| Chess Puzzles | 6% | 1% |
| LMArena Hard Prompts | 1384 | 1220 |
| Epoch Capabilities Index | 136.78 | 130.42 |
| ARC-AGI-2 | 0.4% | — |
| SimpleBench | 27% | — |
| Kagi LLM Benchmark | 52.3% | — |
| ARC-AGI-1 | 5.5% | — |
| EnigmaEval | 2.2% | — |
| LiveBench Reasoning | — | 47.8% |
| DTBench | 68.3% | — |
| LiveBench Data Analysis | — | 45.2% |
| LMCA | 25.6% | — |
| ForecastBench | 61.5 | — |
| LiveBench | — | 41.6% |
Math GPT-4.1 leads
GPT-4.1: 22.3 (#280), Phi-4: 20.8 (#285)
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 38.3% | 13.8% |
| LMArena Math | 1370 | 1246 |
| MATH Level 5 | 83% | 64.9% |
| FrontierMath (Tiers 1-3) | 6% | — |
| Omni-MATH | 47.1% | — |
| LiveBench Math | — | 42% |
| FrontierMath (Feb 2025 set) | 5.5% | — |
| FrontierMath Tier 4 (v1) | 0% | — |
Knowledge GPT-4.1 leads
GPT-4.1: 37.1 (#160), Phi-4: 32.6 (#209)
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| GPQA Diamond | 66.9% | 56.1% |
| Vectara Hallucination Rate | 5.6% | 3.7% |
| LMArena Expert | 1364 | 1203 |
| Humanity's Last Exam | 5.4% | — |
| SimpleQA Verified | 31.1% | — |
| MMLU-Pro | 81.1% | — |
| Confabulations | — | 29.4% |
| GPQA (HELM) | 65.9% | — |
| MMLU | — | 84.8% |
Multimodal Not comparable
GPT-4.1: 38.2 (#67), Phi-4: —
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| LMArena Vision | 1211 | — |
| GeoBench | 72% | — |
Multilingual GPT-4.1 leads
GPT-4.1: 49.4 (#133), Phi-4: 37.2 (#237)
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| LMArena Non-English | 1370 | 1197 |
| LMArena Chinese | 1382 | 1212 |
| LMArena French | 1382 | 1224 |
| LMArena German | 1381 | 1222 |
| LMArena Japanese | 1319 | 1158 |
| LMArena Korean | 1339 | 1151 |
| LMArena Russian | 1377 | 1209 |
| LMArena Spanish | 1376 | 1234 |
Instruction Following GPT-4.1 leads
GPT-4.1: 71.3 (#153), Phi-4: 60.4 (#251)
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1367 | 1201 |
| LiveBench Instruction Following | — | 58.4% |
| IFEval | 83.8% | — |
Long Context GPT-4.1 leads
GPT-4.1: 40.0 (#163), Phi-4: 36.9 (#226)
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1385 | 1217 |
| Fiction.LiveBench | 63.9% | — |
Writing & Preference GPT-4.1 leads
GPT-4.1: 57.6 (#125), Phi-4: 40.5 (#244)
| Benchmark | GPT-4.1 | Phi-4 |
|---|---|---|
| LMArena Text | 1383 | 1217 |
| LMArena Creative Writing | 1363 | 1182 |
| LMArena Multi-Turn | 1398 | 1206 |
| Short-Story Creative Writing | — | 62.6% |
| EQ-Bench Creative Writing | 1420 | — |
| WildBench | 85.4% | — |
| LiveBench Language | — | 25.6% |
Frequently asked questions
Is GPT-4.1 better than Phi-4?
GPT-4.1 is the stronger model overall, scoring 35.9 to 31.2 on the Noometry Index. Phi-4 costs 40× less per token, which makes it the better buy when GPT-4.1's lead doesn't matter for your workload.
Which is cheaper, GPT-4.1 or Phi-4?
Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; GPT-4.1 lists at $2 and $8.
Is GPT-4.1 or Phi-4 better for coding?
They score almost the same on coding (34.4 vs 34.4); test both on your own repository before choosing.
Which has the bigger context window?
GPT-4.1 does, with 1.05M tokens against 128K.
How many benchmarks do GPT-4.1 and Phi-4 share?
24 benchmarks have published results for both models. GPT-4.1 has 52 scored results on Noometry and Phi-4 has 37.