Model comparison
GPT-5.4 vs Phi-4
GPT-5.4 is the stronger model overall, scoring 59.4 to 31.2 on the Noometry Index. Phi-4 costs 64× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Last verified . 22 shared benchmarks.
MicrosoftSummary
- They share 22 benchmarks with published results for both. GPT-5.4 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 GPT-5.4 leads 73.5 to 20.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 97.8% for GPT-5.4 and 13.8% for Phi-4.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
- GPT-5.4 accepts more context: 1.05M tokens versus 128K.
- Phi-4 has downloadable open weights; the other is API-only.
Side by side
| GPT-5.4 | Phi-4 | |
|---|---|---|
| Provider | OpenAI | Microsoft |
| Noometry Index | 59.4 | 31.2 |
| Released | 2026-03-05 | 2024-12-11 |
| Weights | Proprietary | Open |
| Context window | 1.05M | 128K |
| Max output | 128K | 4K |
| Input $ / M tokens | $2.50 | $0.07 |
| Output $ / M tokens | $15 | $0.14 |
| Results tracked | 68 | 37 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.4 leads
GPT-5.4: 52.6 (#33), Phi-4: 34.4 (#239)
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| LMArena Coding | 1497 | 1231 |
| SWE-bench Verified | 76.9% | — |
| DeepSWE | 51.8% | — |
| LMArena WebDev | 1465 | — |
| SciCode | 56.6% | — |
| GSO | 31.4% | — |
| WeirdML | 77.7% | — |
| BigCodeBench Instruct | — | 45.5% |
| LiveBench Coding | — | 30.7% |
| MirrorCode | 15.6% | — |
| BigCodeBench Complete | — | 55.4% |
| ALE-Bench | 1,607 | — |
| AlgoTune | 1.85 | — |
Agentic & Tool Use GPT-5.4 leads
GPT-5.4: 46.5 (#13), Phi-4: 22.8 (#128)
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| Terminal-Bench | 81.8% | — |
| APEX-Agents | 52.4% | — |
| Berkeley Function Calling Leaderboard | — | 28.8% |
| τ²-bench Banking | 39.4% | — |
| DeepResearch Bench | 35.1% | — |
| PostTrainBench | 19% | — |
| BALROG | — | 11.6% |
| GBAEval | 45.1% | — |
| LMArena Search | 1197 | — |
| METR Time Horizons | 74.3% | — |
| Vending-Bench 2 | 6,144 | — |
Reasoning GPT-5.4 leads
GPT-5.4: 61.8 (#19), Phi-4: 17.7 (#291)
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| Chess Puzzles | 44% | 1% |
| LMArena Hard Prompts | 1485 | 1220 |
| Epoch Capabilities Index | 156.81 | 130.42 |
| ARC-AGI-2 | 74% | — |
| Kagi LLM Benchmark | 63.8% | — |
| NYT Connections (extended) | 91.3% | — |
| ARC-AGI-1 | 93.7% | — |
| CritPt | 23.4% | — |
| EnigmaEval | 16% | — |
| Thematic Generalization | 80% | — |
| EBR-Bench | 25.4% | — |
| LiveBench Reasoning | — | 47.8% |
| Mystery Game Puzzles | 37% | — |
| DTBench | 94.4% | — |
| LiveBench Data Analysis | — | 45.2% |
| LMCA | 52% | — |
| ForecastBench | 59.5 | — |
| LiveBench | — | 41.6% |
Math GPT-5.4 leads
GPT-5.4: 73.5 (#19), Phi-4: 20.8 (#285)
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 97.8% | 13.8% |
| LMArena Math | 1488 | 1246 |
| FrontierMath (Tiers 1-3) | 78.6% | — |
| FrontierMath Tier 4 | 49% | — |
| MathArena Final-Answer Competitions | 83.1% | — |
| ProofBench | 56% | — |
| LiveBench Math | — | 42% |
| MATH Level 5 | — | 64.9% |
| FrontierMath (Feb 2025 set) | 47.6% | — |
| FrontierMath Tier 4 (v1) | 27.1% | — |
Knowledge GPT-5.4 leads
GPT-5.4: 65.3 (#14), Phi-4: 32.6 (#209)
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| GPQA Diamond | 93.3% | 56.1% |
| Vectara Hallucination Rate | 7% | 3.7% |
| LMArena Expert | 1507 | 1203 |
| Humanity's Last Exam | 36.2% | — |
| SimpleQA Verified | 45.1% | — |
| Confabulations | — | 29.4% |
| MMLU | — | 84.8% |
Multimodal Not comparable
GPT-5.4: 43.7 (#20), Phi-4: —
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| LMArena Vision | 1303 | — |
| Blueprint-Bench 2 | 27.1% | — |
| Furniture Assembly | 37.5% | — |
| LMArena Document | 1471 | — |
Multilingual GPT-5.4 leads
GPT-5.4: 56.2 (#23), Phi-4: 37.2 (#237)
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| LMArena Non-English | 1465 | 1197 |
| LMArena Chinese | 1519 | 1212 |
| LMArena French | 1493 | 1224 |
| LMArena German | 1472 | 1222 |
| LMArena Japanese | 1485 | 1158 |
| LMArena Korean | 1448 | 1151 |
| LMArena Russian | 1480 | 1209 |
| LMArena Spanish | 1454 | 1234 |
Instruction Following GPT-5.4 leads
GPT-5.4: 77.1 (#27), Phi-4: 60.4 (#251)
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1469 | 1201 |
| LiveBench Instruction Following | — | 58.4% |
Long Context GPT-5.4 leads
GPT-5.4: 50.3 (#8), Phi-4: 36.9 (#226)
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1473 | 1217 |
| CL-bench | 27.9% | — |
| CL-bench Life | 21.7% | — |
Writing & Preference GPT-5.4 leads
GPT-5.4: 71.9 (#17), Phi-4: 40.5 (#244)
| Benchmark | GPT-5.4 | Phi-4 |
|---|---|---|
| LMArena Text | 1469 | 1217 |
| LMArena Creative Writing | 1439 | 1182 |
| LMArena Multi-Turn | 1482 | 1206 |
| Short-Story Creative Writing | — | 62.6% |
| EQ-Bench Creative Writing | 1840 | — |
| EQ-Bench 4 | 1272 | — |
| LiveBench Language | — | 25.6% |
Frequently asked questions
Is GPT-5.4 better than Phi-4?
GPT-5.4 is the stronger model overall, scoring 59.4 to 31.2 on the Noometry Index. Phi-4 costs 64× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Which is cheaper, GPT-5.4 or Phi-4?
Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; GPT-5.4 lists at $2.50 and $15.
Is GPT-5.4 or Phi-4 better for coding?
GPT-5.4 scores higher on coding benchmarks: 52.6 versus 34.4 in the Noometry coding category.
Which has the bigger context window?
GPT-5.4 does, with 1.05M tokens against 128K.
How many benchmarks do GPT-5.4 and Phi-4 share?
22 benchmarks have published results for both models. GPT-5.4 has 68 scored results on Noometry and Phi-4 has 37.