GPT-5 Nano vs Phi 4
Updated October 21, 2026. 41 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Neither model has a coding score yet, so the two cannot be ranked on coding.
GPT-5 Nano has the edge where the context window is larger, 400K against 16K tokens.
- Shared results
- 41
- GPT-5 Nano only
- 44
- Phi 4 only
- 4
- Like-for-like categories
- 7 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
GPT-5 Nano
GPT-5 Nano leads, 13.0 against 5.9.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only GPT-5 Nano has a GPQA Diamond result, 63.4%.
Directional only
Math
AIME
No call
Only GPT-5 Nano has a AIME result, 81.2%.
Directional only
Knowledge
MMLU Pro
No call
Only GPT-5 Nano has a MMLU Pro result, 76.1%.
Directional only
Web design
Design Arena: websites
No call
Only GPT-5 Nano has a Design Arena: websites result, 1101.
Directional only
Long documents
Listed context window
GPT-5 Nano
GPT-5 Nano takes 400K tokens against 16K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Phi 4
GPT-5 Nano $0.0003, Phi 4 $0.0001 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GPT-5 Nano has the number, $0.109.
Directional only
At a glance
- Coding index
- GPT-5 Nano not benchmarked, Phi 4 not benchmarked.
- Widest category
- Math, 42.2 points apart on average over 6 shared benchmarks.
- Widest single gap
- OTIS Mock AIME: GPT-5 Nano 81.1% against Phi 4 13.8%, 67.4 points apart on a 0 to 100 scale.
- Output price
- GPT-5 Nano $0.40 against Phi 4 $0.14 per million output tokens, 2.9x apart.
- Context window
- GPT-5 Nano takes 400K tokens, Phi 4 16K.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | GPT-5 Nano | Bars | Phi 4 | Ties |
|---|---|---|---|---|
| Coding | 4 | 0 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 6 | 0 | ||
| Math | 6 | 0 | ||
| Knowledge | 5 | 0 | ||
| General | 17 | 0 | ||
| All shared boards | 41 | 0 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index40
| Benchmark | GPT-5 Nano | Phi 4 | Leader |
|---|---|---|---|
| SWE-bench Verified | 34.8% | Not measured | Only GPT-5 Nano measured |
| LiveCodeBench (Vals AI) | 70.2% | Not measured | Only GPT-5 Nano measured |
| LiveCodeBench (Artificial Analysis run) | 78.9% | 23.1% | GPT-5 Nano |
| Terminal-Bench Hard (Artificial Analysis run) | 17.4% | 3.8% | GPT-5 Nano |
| Terminal-Bench | 21.8% | Not measured | Only GPT-5 Nano measured |
| WeirdML | 38.1% | Not measured | Only GPT-5 Nano measured |
| LMArena Text: coding | 1351 | 1231 | GPT-5 Nano |
| LMArena Text: software and IT | 1345 | 1228 | GPT-5 Nano |
| Neither model has a result on the other 30 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | GPT-5 Nano | Phi 4 | Leader |
|---|---|---|---|
| AA Intelligence Index | 13.0 | 5.9 | GPT-5 Nano |
| Epoch Capabilities Index | 139.2 | 130.5 | GPT-5 Nano |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | GPT-5 Nano | Phi 4 | Leader |
|---|---|---|---|
| τ²-Bench Telecom | 36.5% | 0.0% | GPT-5 Nano |
| BALROG | Not measured | 11.6% | Only Phi 4 measured |
| Neither model has a result on the other 37 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | GPT-5 Nano | Phi 4 | Leader |
|---|---|---|---|
| GPQA Diamond | 63.4% | Not measured | Only GPT-5 Nano measured |
| MMMU Pro (Vals AI) | 70.9% | Not measured | Only GPT-5 Nano measured |
| GPQA Diamond (Artificial Analysis run) | 67.6% | 57.5% | GPT-5 Nano |
| Humanity's Last Exam (Artificial Analysis run) | 9.5% | 3.8% | GPT-5 Nano |
| GPQA Diamond (Epoch AI run) | 69.4% | 56.1% | GPT-5 Nano |
| ARC-AGI | 20.7% | Not measured | Only GPT-5 Nano measured |
| ARC-AGI-2 | 2.6% | Not measured | Only GPT-5 Nano measured |
| Chess puzzles | 22.0% | Not measured | Only GPT-5 Nano measured |
| Mystery game puzzles | 9.0% | Not measured | Only GPT-5 Nano measured |
| VPCT | 37.2% | Not measured | Only GPT-5 Nano measured |
| LMArena Text: hard prompts | 1328 | 1220 | GPT-5 Nano |
| LMArena Text: expert | 1321 | 1203 | GPT-5 Nano |
| LMArena Text: hard prompts in English | 1338 | 1230 | GPT-5 Nano |
| LMArena Vision: diagrams | 1141 | Not measured | Only GPT-5 Nano measured |
| LMArena Vision: homework | 1157 | Not measured | Only GPT-5 Nano measured |
| Neither model has a result on the other 6 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | GPT-5 Nano | Phi 4 | Leader |
|---|---|---|---|
| AIME | 81.2% | Not measured | Only GPT-5 Nano measured |
| MATH 500 (Vals AI) | 93.8% | Not measured | Only GPT-5 Nano measured |
| MGSM (Vals AI) | 89.3% | Not measured | Only GPT-5 Nano measured |
| AA Math Index | 83.7 | 18.0 | GPT-5 Nano |
| AIME 2025 (Artificial Analysis run) | 83.7% | 18.0% | GPT-5 Nano |
| OTIS Mock AIME | 81.1% | 13.8% | GPT-5 Nano |
| FrontierMath, tiers 1 to 3 | 20.0% | Not measured | Only GPT-5 Nano measured |
| FrontierMath, tier 4 | 2.4% | Not measured | Only GPT-5 Nano measured |
| ProofBench | 12.0% | Not measured | Only GPT-5 Nano measured |
| MATH level 5 | 95.2% | 64.9% | GPT-5 Nano |
| LMArena Text: math | 1317 | 1246 | GPT-5 Nano |
| LMArena Text: mathematical work | 1326 | 1254 | GPT-5 Nano |
| Neither model has a result on the other 1 board here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | GPT-5 Nano | Phi 4 | Leader |
|---|---|---|---|
| MMLU Pro | 76.1% | Not measured | Only GPT-5 Nano measured |
| LegalBench (Vals AI) | 50.1% | Not measured | Only GPT-5 Nano measured |
| CaseLaw (Vals AI) | 52.6% | Not measured | Only GPT-5 Nano measured |
| TaxEval (Vals AI) | 67.4% | Not measured | Only GPT-5 Nano measured |
| MortgageTax (Vals AI) | 53.6% | Not measured | Only GPT-5 Nano measured |
| MedQA (Vals AI) | 93.3% | Not measured | Only GPT-5 Nano measured |
| MedCode (Vals AI) | 30.4% | Not measured | Only GPT-5 Nano measured |
| MMLU Pro (Artificial Analysis run) | 78.0% | 71.4% | GPT-5 Nano |
| SimpleQA Verified | 11.7% | Not measured | Only GPT-5 Nano measured |
| MMLU | Not measured | 84.8% | Only Phi 4 measured |
| LMArena Text: science | 1318 | 1206 | GPT-5 Nano |
| LMArena Text: medicine and healthcare | 1335 | 1202 | GPT-5 Nano |
| LMArena Text: legal and government | 1298 | 1241 | GPT-5 Nano |
| LMArena Text: business and finance | 1311 | 1199 | GPT-5 Nano |
| Neither model has a result on the other 3 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | GPT-5 Nano | Phi 4 | Leader |
|---|---|---|---|
| Design Arena: websites | 1101 | Not measured | Only GPT-5 Nano measured |
| Design Arena: UI components | 1067 | Not measured | Only GPT-5 Nano measured |
| Design Arena: game dev | 1045 | Not measured | Only GPT-5 Nano measured |
| Design Arena: data viz | 1056 | Not measured | Only GPT-5 Nano measured |
| Design Arena: 3D | 966 | Not measured | Only GPT-5 Nano measured |
| Design Arena: all code categories | 1084 | Not measured | Only GPT-5 Nano measured |
| Neither model has a result on the other 3 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index170
| Benchmark | GPT-5 Nano | Phi 4 | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 30.4% | Not measured | Only GPT-5 Nano measured |
| MedScribe (Vals AI) | 72.9% | Not measured | Only GPT-5 Nano measured |
| IFBench | 67.6% | 23.5% | GPT-5 Nano |
| Long context reasoning (AA-LCR) | 45.0% | 0.0% | GPT-5 Nano |
| DTBench | 62.7% | Not measured | Only GPT-5 Nano measured |
| LMCA | 7.9% | Not measured | Only GPT-5 Nano measured |
| Fiction.LiveBench | 44.4% | Not measured | Only GPT-5 Nano measured |
| Creative writing (Lech Mazur) | Not measured | 62.6% | Only Phi 4 measured |
| LMArena Text | 1320 | 1217 | GPT-5 Nano |
| LMArena Text: instruction following | 1306 | 1201 | GPT-5 Nano |
| LMArena Text: creative writing | 1249 | 1181 | GPT-5 Nano |
| LMArena Vision | 1159 | Not measured | Only GPT-5 Nano measured |
| LMArena Text: English | 1321 | 1231 | GPT-5 Nano |
| LMArena Text: non-English | 1313 | 1197 | GPT-5 Nano |
| LMArena Text: Chinese | 1356 | 1212 | GPT-5 Nano |
| LMArena Text: German | 1327 | 1222 | GPT-5 Nano |
| LMArena Text: Japanese | 1227 | 1158 | GPT-5 Nano |
| LMArena Text: Korean | 1269 | 1151 | GPT-5 Nano |
| LMArena Text: French | Not measured | 1224 | Only Phi 4 measured |
| LMArena Text: Spanish | 1360 | 1234 | GPT-5 Nano |
| LMArena Text: Russian | 1296 | 1209 | GPT-5 Nano |
| LMArena Text: Polish | 1314 | Not measured | Only GPT-5 Nano measured |
| LMArena Text: multi-turn | 1311 | 1206 | GPT-5 Nano |
| LMArena Text: longer queries | 1312 | 1217 | GPT-5 Nano |
| LMArena Text: writing and language | 1283 | 1198 | GPT-5 Nano |
| LMArena Text: entertainment and media | 1268 | 1162 | GPT-5 Nano |
| LMArena Vision: English | 1170 | Not measured | Only GPT-5 Nano measured |
| LMArena Vision: Chinese | 1176 | Not measured | Only GPT-5 Nano measured |
| LMArena Vision: reading text | 1152 | Not measured | Only GPT-5 Nano measured |
| LMArena Vision: humor | 1110 | Not measured | Only GPT-5 Nano measured |
| Neither model has a result on the other 16 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5 Nano | Phi 4 | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Phi 4 costs less
Repository review
50K input + 3K output tokens
GPT-5 Nano costs less
Agent loop
200K input + 10K output tokens
GPT-5 Nano costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | OpenAI GPT-5 Nano | Microsoft Phi 4 |
|---|---|---|
| Maker | OpenAI | Microsoft |
| Weights | No public weights listed | Open weights |
| Released | August 7, 2025 | January 10, 2025 |
| Context window | 400K | 16K |
| Input $/Mtok | $0.05 | $0.07 |
| Output $/Mtok | $0.40 | $0.14 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 0 |
| Cost per solved task | $0.109 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5 Nano better than Phi 4 for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. GPT-5 Nano has the edge where the context window is larger, 400K against 16K tokens.
Which is cheaper, GPT-5 Nano or Phi 4?
GPT-5 Nano costs $0.05 per million input tokens and $0.40 per million output tokens; Phi 4 costs $0.07 and $0.14. Cost per solved task is known for GPT-5 Nano only.
Which has the bigger context window, GPT-5 Nano or Phi 4?
GPT-5 Nano, with 400K tokens against 16K.
How many benchmarks is the GPT-5 Nano vs Phi 4 comparison based on?
41 benchmarks have a result for both models as of October 21, 2026. GPT-5 Nano's coding index rests on 0 independent benchmarks and Phi 4's on 0. Where only one model has a result, the page says so instead of naming a winner.