Llama 4 Scout vs Phi 4
Updated October 21, 2026. 42 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.
Llama 4 Scout has the edge where the context window is larger, 1.31M against 16K tokens, and where what matters is Terminal-Bench Hard (Artificial Analysis run) or Epoch Capabilities Index, where it leads.
- Shared results
- 42
- Llama 4 Scout only
- 38
- Phi 4 only
- 3
- 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
Llama 4 Scout
Llama 4 Scout leads, 6.5 against 5.9.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Llama 4 Scout has a AA Agentic Index result, 0.5.
Directional only
Reasoning
GPQA Diamond
No call
Only Llama 4 Scout has a GPQA Diamond result, 47.0%.
Directional only
Math
AIME
No call
Only Llama 4 Scout has a AIME result, 19.0%.
Directional only
Knowledge
MMLU Pro
No call
Only Llama 4 Scout has a MMLU Pro result, 69.6%.
Directional only
Web design
Design Arena: websites
No call
Only Llama 4 Scout has a Design Arena: websites result, 750.
Directional only
Long documents
Listed context window
Llama 4 Scout
Llama 4 Scout takes 1.31M tokens against 16K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Phi 4
Llama 4 Scout $0.0004, Phi 4 $0.0001 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Llama 4 Scout has the number, $1.276.
Directional only
At a glance
- Coding index
- Llama 4 Scout not benchmarked, Phi 4 not benchmarked.
- Widest category
- Agentic work, 15.5 points apart on average over 1 shared benchmark.
- Widest single gap
- Long context reasoning (AA-LCR): Llama 4 Scout 27.7% against Phi 4 0.0%, 27.7 points apart on a 0 to 100 scale.
- Output price
- Llama 4 Scout $0.50 against Phi 4 $0.14 per million output tokens, 3.6x apart.
- Context window
- Llama 4 Scout takes 1.31M 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 | Llama 4 Scout | Bars | Phi 4 | Ties |
|---|---|---|---|---|
| Coding | 3 | 1 | ||
| Intelligence | 1 | 1 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 4 | 1 | 1 tie | |
| Math | 2 | 4 | ||
| Knowledge | 5 | 0 | ||
| General | 18 | 0 | ||
| All shared boards | 34 | 7 | 1 tie |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index31
| Benchmark | Llama 4 Scout | Phi 4 | Leader |
|---|---|---|---|
| SWE-bench Verified | 9.1% | Not measured | Only Llama 4 Scout measured |
| AA Coding Index | 8.2 | Not measured | Only Llama 4 Scout measured |
| LiveCodeBench (Vals AI) | 38.5% | Not measured | Only Llama 4 Scout measured |
| LiveCodeBench (Artificial Analysis run) | 29.9% | 23.1% | Llama 4 Scout |
| SciCode | 21.3% | Not measured | Only Llama 4 Scout measured |
| Terminal-Bench Hard (Artificial Analysis run) | 1.5% | 3.8% | Phi 4 |
| LMArena Text: coding | 1286 | 1231 | Llama 4 Scout |
| LMArena Text: software and IT | 1286 | 1228 | Llama 4 Scout |
| Neither model has a result on the other 30 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index11
| Benchmark | Llama 4 Scout | Phi 4 | Leader |
|---|---|---|---|
| AA Intelligence Index | 6.5 | 5.9 | Llama 4 Scout |
| Epoch Capabilities Index | 129.6 | 130.5 | Phi 4 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Llama 4 Scout | Phi 4 | Leader |
|---|---|---|---|
| AA Agentic Index | 0.5 | Not measured | Only Llama 4 Scout measured |
| τ²-Bench Telecom | 15.5% | 0.0% | Llama 4 Scout |
| BALROG | Not measured | 11.6% | Only Phi 4 measured |
| Neither model has a result on the other 36 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index41
| Benchmark | Llama 4 Scout | Phi 4 | Leader |
|---|---|---|---|
| GPQA Diamond | 47.0% | Not measured | Only Llama 4 Scout measured |
| MMMU Pro (Vals AI) | 58.8% | Not measured | Only Llama 4 Scout measured |
| GPQA Diamond (Artificial Analysis run) | 58.7% | 57.5% | Llama 4 Scout |
| Humanity's Last Exam (Artificial Analysis run) | 3.8% | 3.8% | Tie |
| GPQA Diamond (Epoch AI run) | 51.8% | 56.1% | Phi 4 |
| ARC-AGI | 0.5% | Not measured | Only Llama 4 Scout measured |
| ARC-AGI-2 | 0.0% | Not measured | Only Llama 4 Scout measured |
| LMArena Text: hard prompts | 1266 | 1220 | Llama 4 Scout |
| LMArena Text: expert | 1235 | 1203 | Llama 4 Scout |
| LMArena Text: hard prompts in English | 1293 | 1230 | Llama 4 Scout |
| LMArena Vision: diagrams | 1125 | Not measured | Only Llama 4 Scout measured |
| LMArena Vision: homework | 1148 | Not measured | Only Llama 4 Scout measured |
| Neither model has a result on the other 9 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index24
| Benchmark | Llama 4 Scout | Phi 4 | Leader |
|---|---|---|---|
| AIME | 19.0% | Not measured | Only Llama 4 Scout measured |
| MATH 500 (Vals AI) | 79.2% | Not measured | Only Llama 4 Scout measured |
| MGSM (Vals AI) | 88.0% | Not measured | Only Llama 4 Scout measured |
| AA Math Index | 14.0 | 18.0 | Phi 4 |
| AIME 2025 (Artificial Analysis run) | 14.0% | 18.0% | Phi 4 |
| OTIS Mock AIME | 7.8% | 13.8% | Phi 4 |
| MATH level 5 | 62.3% | 64.9% | Phi 4 |
| LMArena Text: math | 1287 | 1246 | Llama 4 Scout |
| LMArena Text: mathematical work | 1286 | 1254 | Llama 4 Scout |
| Neither model has a result on the other 4 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | Llama 4 Scout | Phi 4 | Leader |
|---|---|---|---|
| MMLU Pro | 69.6% | Not measured | Only Llama 4 Scout measured |
| LegalBench (Vals AI) | 72.0% | Not measured | Only Llama 4 Scout measured |
| CorpFin (Vals AI) | 46.8% | Not measured | Only Llama 4 Scout measured |
| TaxEval (Vals AI) | 55.2% | Not measured | Only Llama 4 Scout measured |
| MortgageTax (Vals AI) | 57.8% | Not measured | Only Llama 4 Scout measured |
| MedQA (Vals AI) | 50.9% | Not measured | Only Llama 4 Scout measured |
| MedCode (Vals AI) | 23.3% | Not measured | Only Llama 4 Scout measured |
| MMLU Pro (Artificial Analysis run) | 75.2% | 71.4% | Llama 4 Scout |
| MMLU | Not measured | 84.8% | Only Phi 4 measured |
| LMArena Text: science | 1278 | 1206 | Llama 4 Scout |
| LMArena Text: medicine and healthcare | 1275 | 1202 | Llama 4 Scout |
| LMArena Text: legal and government | 1285 | 1241 | Llama 4 Scout |
| LMArena Text: business and finance | 1257 | 1199 | Llama 4 Scout |
| Neither model has a result on the other 4 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | Llama 4 Scout | Phi 4 | Leader |
|---|---|---|---|
| Design Arena: websites | 750 | Not measured | Only Llama 4 Scout measured |
| Design Arena: UI components | 768 | Not measured | Only Llama 4 Scout measured |
| Design Arena: game dev | 773 | Not measured | Only Llama 4 Scout measured |
| Design Arena: data viz | 896 | Not measured | Only Llama 4 Scout measured |
| Design Arena: all code categories | 790 | Not measured | Only Llama 4 Scout measured |
| Neither model has a result on the other 4 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index180
| Benchmark | Llama 4 Scout | Phi 4 | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 34.8% | Not measured | Only Llama 4 Scout measured |
| MedScribe (Vals AI) | 50.6% | Not measured | Only Llama 4 Scout measured |
| IFBench | 39.5% | 23.5% | Llama 4 Scout |
| Long context reasoning (AA-LCR) | 27.7% | 0.0% | Llama 4 Scout |
| DTBench | 57.9% | Not measured | Only Llama 4 Scout measured |
| LMCA | 12.0% | Not measured | Only Llama 4 Scout measured |
| Fiction.LiveBench | 36.0% | Not measured | Only Llama 4 Scout measured |
| Creative writing (Lech Mazur) | Not measured | 62.6% | Only Phi 4 measured |
| LMArena Text | 1279 | 1217 | Llama 4 Scout |
| LMArena Text: instruction following | 1248 | 1201 | Llama 4 Scout |
| LMArena Text: creative writing | 1249 | 1181 | Llama 4 Scout |
| LMArena Vision | 1118 | Not measured | Only Llama 4 Scout measured |
| LMArena Text: English | 1304 | 1231 | Llama 4 Scout |
| LMArena Text: non-English | 1252 | 1197 | Llama 4 Scout |
| LMArena Text: Chinese | 1255 | 1212 | Llama 4 Scout |
| LMArena Text: German | 1273 | 1222 | Llama 4 Scout |
| LMArena Text: Japanese | 1206 | 1158 | Llama 4 Scout |
| LMArena Text: Korean | 1207 | 1151 | Llama 4 Scout |
| LMArena Text: French | 1283 | 1224 | Llama 4 Scout |
| LMArena Text: Spanish | 1278 | 1234 | Llama 4 Scout |
| LMArena Text: Russian | 1263 | 1209 | Llama 4 Scout |
| LMArena Text: Polish | 1265 | Not measured | Only Llama 4 Scout measured |
| LMArena Text: multi-turn | 1280 | 1206 | Llama 4 Scout |
| LMArena Text: longer queries | 1265 | 1217 | Llama 4 Scout |
| LMArena Text: writing and language | 1269 | 1198 | Llama 4 Scout |
| LMArena Text: entertainment and media | 1242 | 1162 | Llama 4 Scout |
| LMArena Vision: English | 1143 | Not measured | Only Llama 4 Scout measured |
| LMArena Vision: Chinese | 1106 | Not measured | Only Llama 4 Scout measured |
| LMArena Vision: reading text | 1117 | Not measured | Only Llama 4 Scout measured |
| LMArena Vision: humor | 1102 | Not measured | Only Llama 4 Scout measured |
| LMArena Vision: creative writing | 1126 | Not measured | Only Llama 4 Scout measured |
| Neither model has a result on the other 15 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Llama 4 Scout | 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
Phi 4 costs less
Agent loop
200K input + 10K output tokens
Phi 4 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 | Meta Llama 4 Scout | Microsoft Phi 4 |
|---|---|---|
| Maker | Meta Llama | Microsoft |
| Weights | Open weights | Open weights |
| Released | April 5, 2025 | January 10, 2025 |
| Context window | 1.31M | 16K |
| Input $/Mtok | $0.175 | $0.07 |
| Output $/Mtok | $0.50 | $0.14 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 0 |
| Cost per solved task | $1.276 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Llama 4 Scout better than Phi 4 for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Llama 4 Scout has the edge where the context window is larger, 1.31M against 16K tokens, and where what matters is Terminal-Bench Hard (Artificial Analysis run) or Epoch Capabilities Index, where it leads.
Which is cheaper, Llama 4 Scout or Phi 4?
Llama 4 Scout costs $0.175 per million input tokens and $0.50 per million output tokens; Phi 4 costs $0.07 and $0.14. Cost per solved task is known for Llama 4 Scout only.
Which has the bigger context window, Llama 4 Scout or Phi 4?
Llama 4 Scout, with 1.31M tokens against 16K.
How many benchmarks is the Llama 4 Scout vs Phi 4 comparison based on?
42 benchmarks have a result for both models as of October 21, 2026. Llama 4 Scout'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.