Phi 4 vs Inkling
Inkling leads Phi 4 on the Artificial Analysis Intelligence Index, 25.0 to 5.9, and Phi 4 costs 28.9x less per output token ($0.14 vs $4.05 per million). Updated October 21, 2026; 34 benchmarks have a result for both.
Phi 4 or Inkling: which to pick
Only Inkling has a coding score, 31.5 on the Vibe Coding Index; Phi 4 has none, so the two cannot be ranked on coding.
Pick Phi 4 for output tokens 28.9x cheaper ($0.14 vs $4.05 per million).
- Shared results
- 34
- Phi 4 only
- 11
- Inkling only
- 68
- Like-for-like categories
- 6 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
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 Inkling has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
Inkling
Inkling leads, 25.0 against 5.9.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Inkling has a AA Agentic Index result, 22.5.
Directional only
Reasoning
GPQA Diamond
No call
Only Inkling has a GPQA Diamond result, 87.1%.
Directional only
Knowledge
MMLU Pro
No call
Only Inkling has a MMLU Pro result, 86.3%.
Directional only
Web design
Design Arena: websites
No call
Only Inkling has a Design Arena: websites result, 1224.
Directional only
General
Vals Index
No call
Only Inkling has a Vals Index result, 29.0%.
Directional only
Long documents
Listed context window
Inkling
Inkling takes 524K tokens against 16K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Phi 4
Phi 4 $0.0001, Inkling $0.0030 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Inkling has the number, $0.76.
Directional only
At a glance
- Coding index
- Phi 4 not benchmarked, Inkling 31.5 from 6 benchmarks.
- Widest category
- Math, 49.5 points apart on average over 3 shared benchmarks.
- Widest single gap
- Long context reasoning (AA-LCR): Phi 4 0.0% against Inkling 77.3%, 77.3 points apart on a 0 to 100 scale.
- Output price
- Phi 4 $0.14 against Inkling $4.05 per million output tokens, 28.9x apart.
- Context window
- Phi 4 takes 16K tokens, Inkling 524K.
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 | Phi 4 | Bars | Inkling | Ties |
|---|---|---|---|---|
| Coding | 0 | 2 | ||
| Intelligence | 0 | 2 | ||
| Reasoning | 0 | 6 | ||
| Math | 0 | 3 | ||
| Knowledge | 0 | 4 | ||
| General | 0 | 17 | ||
| All shared boards | 0 | 34 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index02
| Benchmark | Phi 4 | Inkling | Leader |
|---|---|---|---|
| LiveBench coding | Not measured | 64.0% | Only Inkling measured |
| LMArena WebDev | Not measured | 1414 | Only Inkling measured |
| AA Coding Index | Not measured | 52.1 | Only Inkling measured |
| LiveCodeBench (Vals AI) | Not measured | 85.5% | Only Inkling measured |
| SWE-bench Verified (Vals AI) | Not measured | 77.6% | Only Inkling measured |
| Terminal-Bench 4 (Vals AI) | Not measured | 0.5% | Only Inkling measured |
| Vibe Code Bench (Vals AI) | Not measured | 19.2% | Only Inkling measured |
| Code migration (Vals AI) | Not measured | 11.8% | Only Inkling measured |
| ProgramBench (Vals AI) | Not measured | 0.0% | Only Inkling measured |
| IOI (Vals AI) | Not measured | 14.9% | Only Inkling measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 47.6% | Only Inkling measured |
| Vibe Code Bench 1-100 (Vals AI) | Not measured | 7.3% | Only Inkling measured |
| SkillsBench (Vals AI) | Not measured | 26.1% | Only Inkling measured |
| CyberBench (Vals AI) | Not measured | 66.8% | Only Inkling measured |
| LiveCodeBench (Artificial Analysis run) | 23.1% | Not measured | Only Phi 4 measured |
| SciCode | Not measured | 47.0% | Only Inkling measured |
| Terminal-Bench Hard (Artificial Analysis run) | 3.8% | Not measured | Only Phi 4 measured |
| FrontierCode | Not measured | 14.0% | Only Inkling measured |
| FrontierSWE | Not measured | 4.1% | Only Inkling measured |
| WeirdML | Not measured | 32.3% | Only Inkling measured |
| ALE-Bench | Not measured | 946.0 | Only Inkling measured |
| LMArena Text: coding | 1231 | 1466 | Inkling |
| LMArena Text: software and IT | 1228 | 1465 | Inkling |
| LMArena WebDev: React | Not measured | 1404 | Only Inkling measured |
| LMArena WebDev: HTML | Not measured | 1399 | Only Inkling measured |
| Neither model has a result on the other 13 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Phi 4 | Inkling | Leader |
|---|---|---|---|
| AA Intelligence Index | 5.9 | 25.0 | Inkling |
| Epoch Capabilities Index | 130.5 | 148.4 | Inkling |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index14 boards, none shared
| Benchmark | Phi 4 | Inkling | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 22.5 | Only Inkling measured |
| Design Arena: agents building web apps | Not measured | 1196 | Only Inkling measured |
| LiveBench research | Not measured | 59.9% | Only Inkling measured |
| LiveBench browser | Not measured | 16.0% | Only Inkling measured |
| LiveBench long agentic | Not measured | 14.1% | Only Inkling measured |
| SREBench (Vals AI) | Not measured | 0.0% | Only Inkling measured |
| Finance Agent (Vals AI) | Not measured | 46.6% | Only Inkling measured |
| Tax Agent Bench (Vals AI) | Not measured | 15.3% | Only Inkling measured |
| Legal Research Bench (Vals AI) | Not measured | 28.4% | Only Inkling measured |
| Harvey's Legal Agent Benchmark (Vals AI) | Not measured | 3.8% | Only Inkling measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 38.4% | Only Inkling measured |
| τ²-Bench Telecom | 0.0% | Not measured | Only Phi 4 measured |
| APEX-Agents | Not measured | 33.8% | Only Inkling measured |
| BALROG | 11.6% | Not measured | Only Phi 4 measured |
| Neither model has a result on the other 25 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index06
| Benchmark | Phi 4 | Inkling | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 87.1% | Only Inkling measured |
| LiveBench reasoning | Not measured | 78.3% | Only Inkling measured |
| MMMU Pro (Vals AI) | Not measured | 78.5% | Only Inkling measured |
| GPQA Diamond (Artificial Analysis run) | 57.5% | 87.2% | Inkling |
| Humanity's Last Exam (Artificial Analysis run) | 3.8% | 31.9% | Inkling |
| GPQA Diamond (Epoch AI run) | 56.1% | 88.3% | Inkling |
| ARC-AGI | Not measured | 79.5% | Only Inkling measured |
| ARC-AGI-2 | Not measured | 36.5% | Only Inkling measured |
| SimpleBench | Not measured | 50.0% | Only Inkling measured |
| Chess puzzles | Not measured | 21.0% | Only Inkling measured |
| LMArena Text: hard prompts | 1220 | 1451 | Inkling |
| LMArena Text: expert | 1203 | 1466 | Inkling |
| LMArena Text: hard prompts in English | 1230 | 1448 | Inkling |
| Neither model has a result on the other 8 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Phi 4 | Inkling | Leader |
|---|---|---|---|
| AA Math Index | 18.0 | Not measured | Only Phi 4 measured |
| AIME 2025 (Artificial Analysis run) | 18.0% | Not measured | Only Phi 4 measured |
| OTIS Mock AIME | 13.8% | 88.9% | Inkling |
| FrontierMath, tiers 1 to 3 | Not measured | 33.3% | Only Inkling measured |
| FrontierMath, tier 4 | Not measured | 4.9% | Only Inkling measured |
| ProofBench | Not measured | 0.0% | Only Inkling measured |
| MATH level 5 | 64.9% | Not measured | Only Phi 4 measured |
| LMArena Text: math | 1246 | 1479 | Inkling |
| LMArena Text: mathematical work | 1254 | 1460 | Inkling |
| Neither model has a result on the other 4 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | Phi 4 | Inkling | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 86.3% | Only Inkling measured |
| LegalBench (Vals AI) | Not measured | 83.0% | Only Inkling measured |
| CorpFin (Vals AI) | Not measured | 68.6% | Only Inkling measured |
| TaxEval (Vals AI) | Not measured | 75.3% | Only Inkling measured |
| MortgageTax (Vals AI) | Not measured | 64.0% | Only Inkling measured |
| MedCode (Vals AI) | Not measured | 41.2% | Only Inkling measured |
| MMLU Pro (Artificial Analysis run) | 71.4% | Not measured | Only Phi 4 measured |
| SimpleQA Verified | Not measured | 40.3% | Only Inkling measured |
| MMLU | 84.8% | Not measured | Only Phi 4 measured |
| LMArena Text: science | 1206 | 1455 | Inkling |
| LMArena Text: medicine and healthcare | 1202 | 1447 | Inkling |
| LMArena Text: legal and government | 1241 | 1448 | Inkling |
| LMArena Text: business and finance | 1199 | 1438 | Inkling |
| Neither model has a result on the other 4 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index9 boards, none shared
| Benchmark | Phi 4 | Inkling | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1224 | Only Inkling measured |
| Design Arena: UI components | Not measured | 1201 | Only Inkling measured |
| Design Arena: game dev | Not measured | 1166 | Only Inkling measured |
| Design Arena: data viz | Not measured | 1159 | Only Inkling measured |
| Design Arena: 3D | Not measured | 1161 | Only Inkling measured |
| Design Arena: all code categories | Not measured | 1206 | Only Inkling measured |
| Design Arena: SVG | Not measured | 1110 | Only Inkling measured |
| Design Arena: ASCII art | Not measured | 1089 | Only Inkling measured |
| LiveBench design | Not measured | 58.3% | Only Inkling measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index017
| Benchmark | Phi 4 | Inkling | Leader |
|---|---|---|---|
| Vals Index | Not measured | 29.0% | Only Inkling measured |
| LiveBench data analysis | Not measured | 72.8% | Only Inkling measured |
| Public Benefits Bench (Vals AI) | Not measured | 58.5% | Only Inkling measured |
| SAGE (Vals AI) | Not measured | 36.5% | Only Inkling measured |
| MedScribe (Vals AI) | Not measured | 85.4% | Only Inkling measured |
| Vals Multimodal Index | Not measured | 49.4% | Only Inkling measured |
| IFBench | 23.5% | Not measured | Only Phi 4 measured |
| Long context reasoning (AA-LCR) | 0.0% | 77.3% | Inkling |
| DTBench | Not measured | 87.5% | Only Inkling measured |
| LMCA | Not measured | 37.6% | Only Inkling measured |
| Creative writing (Lech Mazur) | 62.6% | Not measured | Only Phi 4 measured |
| LMArena Text | 1217 | 1442 | Inkling |
| LMArena Text: instruction following | 1201 | 1425 | Inkling |
| LMArena Text: creative writing | 1181 | 1390 | Inkling |
| LMArena Text: English | 1231 | 1442 | Inkling |
| LMArena Text: non-English | 1197 | 1434 | Inkling |
| LMArena Text: Chinese | 1212 | 1493 | Inkling |
| LMArena Text: German | 1222 | 1445 | Inkling |
| LMArena Text: Japanese | 1158 | 1429 | Inkling |
| LMArena Text: Korean | 1151 | 1403 | Inkling |
| LMArena Text: French | 1224 | 1455 | Inkling |
| LMArena Text: Spanish | 1234 | 1449 | Inkling |
| LMArena Text: Russian | 1209 | 1430 | Inkling |
| LMArena Text: Polish | Not measured | 1450 | Only Inkling measured |
| LMArena Text: multi-turn | 1206 | 1437 | Inkling |
| LMArena Text: longer queries | 1217 | 1435 | Inkling |
| LMArena Text: writing and language | 1198 | 1404 | Inkling |
| LMArena Text: entertainment and media | 1162 | 1394 | Inkling |
| Neither model has a result on the other 18 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Phi 4 | Inkling | 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 | Microsoft Phi 4 | Thinking Machines Inkling |
|---|---|---|
| Maker | Microsoft | Thinking Machines |
| Weights | Open weights | Open weights |
| Released | January 10, 2025 | July 17, 2026 |
| Context window | 16K | 524K |
| Input $/Mtok | $0.07 | $0.975 |
| Output $/Mtok | $0.14 | $4.05 |
| Coding index | Not benchmarked | 31.5 |
| Independent measurements behind it | 0 | 6 |
| Cost per solved task | Not computable | $0.76 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Phi 4 better than Inkling for coding?
Only Inkling has a coding score, 31.5 on the Vibe Coding Index; Phi 4 has none, so the two cannot be ranked on coding. Pick Phi 4 for output tokens 28.9x cheaper ($0.14 vs $4.05 per million).
Which is cheaper, Phi 4 or Inkling?
Phi 4 costs $0.07 per million input tokens and $0.14 per million output tokens; Inkling costs $0.975 and $4.05. Cost per solved task is known for Inkling only.
Which has the bigger context window, Phi 4 or Inkling?
Inkling, with 524K tokens against 16K.
How many benchmarks is the Phi 4 vs Inkling comparison based on?
34 benchmarks have a result for both models as of October 21, 2026. Phi 4's coding index rests on 0 independent benchmarks and Inkling's on 6. Where only one model has a result, the page says so instead of naming a winner.