Kimi K2.5 vs Phi 4
Updated October 21, 2026. 35 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Only Kimi K2.5 has a coding score, 38.6 on the Vibe Coding Index; Phi 4 has none, so the two cannot be ranked on coding.
Phi 4 has the edge where output tokens cost 20.4x less.
- Shared results
- 35
- Kimi K2.5 only
- 63
- Phi 4 only
- 10
- 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
Only Kimi K2.5 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
Kimi K2.5
Kimi K2.5 leads, 23.5 against 5.9.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only Kimi K2.5 has a GPQA Diamond result, 84.1%.
Directional only
Math
AIME
No call
Only Kimi K2.5 has a AIME result, 95.6%.
Directional only
Knowledge
MMLU Pro
No call
Only Kimi K2.5 has a MMLU Pro result, 85.9%.
Directional only
Web design
Design Arena: websites
No call
Only Kimi K2.5 has a Design Arena: websites result, 1249.
Directional only
Long documents
Listed context window
Kimi K2.5
Kimi K2.5 takes 262K tokens against 16K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Phi 4
Kimi K2.5 $0.0020, Phi 4 $0.0001 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Kimi K2.5 has the number, $0.24.
Directional only
At a glance
- Coding index
- Kimi K2.5 38.6 from 3 benchmarks, Phi 4 not benchmarked.
- Widest category
- Agentic work, 95.9 points apart on average over 1 shared benchmark.
- Widest single gap
- τ²-Bench Telecom: Kimi K2.5 95.9% against Phi 4 0.0%, 95.9 points apart on a 0 to 100 scale.
- Output price
- Kimi K2.5 $2.85 against Phi 4 $0.14 per million output tokens, 20.4x apart.
- Context window
- Kimi K2.5 takes 262K 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 | Kimi K2.5 | Bars | Phi 4 | Ties |
|---|---|---|---|---|
| Coding | 3 | 0 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 5 | 0 | ||
| Math | 2 | 0 | ||
| Knowledge | 4 | 0 | ||
| General | 18 | 0 | ||
| All shared boards | 35 | 0 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index30
| Benchmark | Kimi K2.5 | Phi 4 | Leader |
|---|---|---|---|
| SWE-bench Verified | 70.8% | Not measured | Only Kimi K2.5 measured |
| SWE-bench Multilingual | 67.3% | Not measured | Only Kimi K2.5 measured |
| LMArena WebDev | 1437 | Not measured | Only Kimi K2.5 measured |
| AA Coding Index | 46.8 | Not measured | Only Kimi K2.5 measured |
| LiveCodeBench (Vals AI) | 83.9% | Not measured | Only Kimi K2.5 measured |
| SWE-bench Verified (Vals AI) | 70.0% | Not measured | Only Kimi K2.5 measured |
| Vibe Code Bench (Vals AI) | 17.5% | Not measured | Only Kimi K2.5 measured |
| Code migration (Vals AI) | 7.0% | Not measured | Only Kimi K2.5 measured |
| Terminal-Bench 2.1 (Vals AI) | 42.0% | Not measured | Only Kimi K2.5 measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 23.1% | Only Phi 4 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 34.8% | 3.8% | Kimi K2.5 |
| Terminal-Bench | 43.2% | Not measured | Only Kimi K2.5 measured |
| WeirdML | 45.6% | Not measured | Only Kimi K2.5 measured |
| ALE-Bench | 821.6 | Not measured | Only Kimi K2.5 measured |
| LMArena Text: coding | 1483 | 1231 | Kimi K2.5 |
| LMArena Text: software and IT | 1471 | 1228 | Kimi K2.5 |
| LMArena WebDev: React | 1429 | Not measured | Only Kimi K2.5 measured |
| LMArena WebDev: HTML | 1426 | Not measured | Only Kimi K2.5 measured |
| LMArena WebDev: from an image | 1439 | Not measured | Only Kimi K2.5 measured |
| Neither model has a result on the other 19 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | Kimi K2.5 | Phi 4 | Leader |
|---|---|---|---|
| AA Intelligence Index | 23.5 | 5.9 | Kimi K2.5 |
| Epoch Capabilities Index | 148.1 | 130.5 | Kimi K2.5 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Kimi K2.5 | Phi 4 | Leader |
|---|---|---|---|
| Design Arena: agents building web apps | 1136 | Not measured | Only Kimi K2.5 measured |
| Design Arena: agents building full-stack apps | 1114 | Not measured | Only Kimi K2.5 measured |
| Design Arena: agents building mobile apps | 1147 | Not measured | Only Kimi K2.5 measured |
| Design Arena: agents building Android apps | 1125 | Not measured | Only Kimi K2.5 measured |
| Design Arena: agents building Godot games | 1219 | Not measured | Only Kimi K2.5 measured |
| Finance Agent (Vals AI) | 35.8% | Not measured | Only Kimi K2.5 measured |
| Legal Research Bench (Vals AI) | 15.9% | Not measured | Only Kimi K2.5 measured |
| Excel Modeling Benchmark (Vals AI) | 28.5% | Not measured | Only Kimi K2.5 measured |
| τ²-Bench Telecom | 95.9% | 0.0% | Kimi K2.5 |
| BALROG | Not measured | 11.6% | Only Phi 4 measured |
| OSWorld | 63.3% | Not measured | Only Kimi K2.5 measured |
| Vending-Bench 2 | $1,198 | Not measured | Only Kimi K2.5 measured |
| Neither model has a result on the other 27 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | Kimi K2.5 | Phi 4 | Leader |
|---|---|---|---|
| GPQA Diamond | 84.1% | Not measured | Only Kimi K2.5 measured |
| MMMU Pro (Vals AI) | 84.3% | Not measured | Only Kimi K2.5 measured |
| GPQA Diamond (Artificial Analysis run) | 87.9% | 57.5% | Kimi K2.5 |
| Humanity's Last Exam (Artificial Analysis run) | 30.7% | 3.8% | Kimi K2.5 |
| GPQA Diamond (Epoch AI run) | Not measured | 56.1% | Only Phi 4 measured |
| ARC-AGI | 65.3% | Not measured | Only Kimi K2.5 measured |
| ARC-AGI-2 | 11.8% | Not measured | Only Kimi K2.5 measured |
| SimpleBench | 46.8% | Not measured | Only Kimi K2.5 measured |
| LMArena Text: hard prompts | 1453 | 1220 | Kimi K2.5 |
| LMArena Text: expert | 1466 | 1203 | Kimi K2.5 |
| LMArena Text: hard prompts in English | 1460 | 1230 | Kimi K2.5 |
| LMArena Vision: diagrams | 1282 | Not measured | Only Kimi K2.5 measured |
| LMArena Vision: homework | 1297 | Not measured | Only Kimi K2.5 measured |
| Neither model has a result on the other 8 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | Kimi K2.5 | Phi 4 | Leader |
|---|---|---|---|
| AIME | 95.6% | Not measured | Only Kimi K2.5 measured |
| AA Math Index | Not measured | 18.0 | Only Phi 4 measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 18.0% | Only Phi 4 measured |
| OTIS Mock AIME | Not measured | 13.8% | Only Phi 4 measured |
| MATH level 5 | Not measured | 64.9% | Only Phi 4 measured |
| LMArena Text: math | 1470 | 1246 | Kimi K2.5 |
| LMArena Text: mathematical work | 1473 | 1254 | Kimi K2.5 |
| Neither model has a result on the other 6 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | Kimi K2.5 | Phi 4 | Leader |
|---|---|---|---|
| MMLU Pro | 85.9% | Not measured | Only Kimi K2.5 measured |
| CaseLaw (Vals AI) | 58.7% | Not measured | Only Kimi K2.5 measured |
| CorpFin (Vals AI) | 68.3% | Not measured | Only Kimi K2.5 measured |
| TaxEval (Vals AI) | 74.2% | Not measured | Only Kimi K2.5 measured |
| MortgageTax (Vals AI) | 66.5% | Not measured | Only Kimi K2.5 measured |
| MedQA (Vals AI) | 94.4% | Not measured | Only Kimi K2.5 measured |
| MedCode (Vals AI) | 39.3% | Not measured | Only Kimi K2.5 measured |
| MMLU Pro (Artificial Analysis run) | Not measured | 71.4% | Only Phi 4 measured |
| SimpleQA Verified | 34.3% | Not measured | Only Kimi K2.5 measured |
| MMLU | Not measured | 84.8% | Only Phi 4 measured |
| LMArena Text: science | 1459 | 1206 | Kimi K2.5 |
| LMArena Text: medicine and healthcare | 1448 | 1202 | Kimi K2.5 |
| LMArena Text: legal and government | 1441 | 1241 | Kimi K2.5 |
| LMArena Text: business and finance | 1427 | 1199 | Kimi K2.5 |
| Neither model has a result on the other 3 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | Kimi K2.5 | Phi 4 | Leader |
|---|---|---|---|
| Design Arena: websites | 1249 | Not measured | Only Kimi K2.5 measured |
| Design Arena: UI components | 1236 | Not measured | Only Kimi K2.5 measured |
| Design Arena: game dev | 1207 | Not measured | Only Kimi K2.5 measured |
| Design Arena: data viz | 1218 | Not measured | Only Kimi K2.5 measured |
| Design Arena: 3D | 1205 | Not measured | Only Kimi K2.5 measured |
| Design Arena: all code categories | 1236 | Not measured | Only Kimi K2.5 measured |
| Design Arena: SVG | 1149 | Not measured | Only Kimi K2.5 measured |
| Design Arena: ASCII art | 1165 | Not measured | Only Kimi K2.5 measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index180
| Benchmark | Kimi K2.5 | Phi 4 | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 49.9% | Not measured | Only Kimi K2.5 measured |
| MedScribe (Vals AI) | 76.4% | Not measured | Only Kimi K2.5 measured |
| IFBench | 70.2% | 23.5% | Kimi K2.5 |
| Long context reasoning (AA-LCR) | 78.0% | 0.0% | Kimi K2.5 |
| Fiction.LiveBench | 86.1% | Not measured | Only Kimi K2.5 measured |
| Creative writing (Lech Mazur) | Not measured | 62.6% | Only Phi 4 measured |
| CL-bench | 19.3% | Not measured | Only Kimi K2.5 measured |
| CL-bench Life | 13.2% | Not measured | Only Kimi K2.5 measured |
| LMArena Text | 1445 | 1217 | Kimi K2.5 |
| LMArena Text: instruction following | 1431 | 1201 | Kimi K2.5 |
| LMArena Text: creative writing | 1422 | 1181 | Kimi K2.5 |
| LMArena Vision | 1269 | Not measured | Only Kimi K2.5 measured |
| LMArena Text: English | 1453 | 1231 | Kimi K2.5 |
| LMArena Text: non-English | 1433 | 1197 | Kimi K2.5 |
| LMArena Text: Chinese | 1494 | 1212 | Kimi K2.5 |
| LMArena Text: German | 1441 | 1222 | Kimi K2.5 |
| LMArena Text: Japanese | 1423 | 1158 | Kimi K2.5 |
| LMArena Text: Korean | 1412 | 1151 | Kimi K2.5 |
| LMArena Text: French | 1454 | 1224 | Kimi K2.5 |
| LMArena Text: Spanish | 1450 | 1234 | Kimi K2.5 |
| LMArena Text: Russian | 1435 | 1209 | Kimi K2.5 |
| LMArena Text: Polish | 1445 | Not measured | Only Kimi K2.5 measured |
| LMArena Text: multi-turn | 1444 | 1206 | Kimi K2.5 |
| LMArena Text: longer queries | 1445 | 1217 | Kimi K2.5 |
| LMArena Text: writing and language | 1425 | 1198 | Kimi K2.5 |
| LMArena Text: entertainment and media | 1417 | 1162 | Kimi K2.5 |
| LMArena Vision: English | 1274 | Not measured | Only Kimi K2.5 measured |
| LMArena Vision: Chinese | 1328 | Not measured | Only Kimi K2.5 measured |
| LMArena Vision: reading text | 1276 | Not measured | Only Kimi K2.5 measured |
| LMArena Vision: humor | 1284 | Not measured | Only Kimi K2.5 measured |
| LMArena Vision: captioning | 1257 | Not measured | Only Kimi K2.5 measured |
| LMArena Vision: recognizing things | 1285 | Not measured | Only Kimi K2.5 measured |
| LMArena Vision: creative writing | 1269 | Not measured | Only Kimi K2.5 measured |
| Neither model has a result on the other 13 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Kimi K2.5 | 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 | MoonshotAI Kimi K2.5 | Microsoft Phi 4 |
|---|---|---|
| Maker | Moonshot AI | Microsoft |
| Weights | Open weights | Open weights |
| Released | January 27, 2026 | January 10, 2025 |
| Context window | 262K | 16K |
| Input $/Mtok | $0.532 | $0.07 |
| Output $/Mtok | $2.85 | $0.14 |
| Coding index | 38.6 | Not benchmarked |
| Independent measurements behind it | 3 | 0 |
| Cost per solved task | $0.24 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Kimi K2.5 better than Phi 4 for coding?
Only Kimi K2.5 has a coding score, 38.6 on the Vibe Coding Index; Phi 4 has none, so the two cannot be ranked on coding. Phi 4 has the edge where output tokens cost 20.4x less.
Which is cheaper, Kimi K2.5 or Phi 4?
Kimi K2.5 costs $0.532 per million input tokens and $2.85 per million output tokens; Phi 4 costs $0.07 and $0.14. Cost per solved task is known for Kimi K2.5 only.
Which has the bigger context window, Kimi K2.5 or Phi 4?
Kimi K2.5, with 262K tokens against 16K.
How many benchmarks is the Kimi K2.5 vs Phi 4 comparison based on?
35 benchmarks have a result for both models as of October 21, 2026. Kimi K2.5's coding index rests on 3 independent benchmarks and Phi 4's on 0. Where only one model has a result, the page says so instead of naming a winner.