Phi 4 vs Qwen3 14B
Updated October 21, 2026. 14 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.
Phi 4 has the edge where output tokens cost 1.6x less, and where what matters is LiveCodeBench (Artificial Analysis run) or Terminal-Bench Hard (Artificial Analysis run), where it leads.
- Shared results
- 14
- Phi 4 only
- 31
- Qwen3 14B 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
Qwen3 14B
Qwen3 14B leads, 6.7 against 5.9.
Same board, both measured
Long documents
Listed context window
Qwen3 14B
Qwen3 14B takes 41K tokens against 16K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Phi 4
Phi 4 $0.0001, Qwen3 14B $0.0002 per turn.
List prices
At a glance
- Coding index
- Phi 4 not benchmarked, Qwen3 14B not benchmarked.
- Widest category
- Math, 44.2 points apart on average over 3 shared benchmarks.
- Widest single gap
- OTIS Mock AIME: Phi 4 13.8% against Qwen3 14B 66.4%, 52.6 points apart on a 0 to 100 scale.
- Output price
- Phi 4 $0.14 against Qwen3 14B $0.23 per million output tokens, 1.6x apart.
- Context window
- Phi 4 takes 16K tokens, Qwen3 14B 41K.
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 | Qwen3 14B | Ties |
|---|---|---|---|---|
| Coding | 0 | 2 | ||
| Intelligence | 0 | 2 | ||
| Agentic work | 0 | 1 | ||
| Reasoning | 0 | 3 | ||
| Math | 0 | 3 | ||
| Knowledge | 0 | 1 | ||
| General | 0 | 1 | 1 tie | |
| All shared boards | 0 | 13 | 1 tie |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index02
| Benchmark | Phi 4 | Qwen3 14B | Leader |
|---|---|---|---|
| LiveCodeBench (Artificial Analysis run) | 23.1% | 52.3% | Qwen3 14B |
| SciCode | Not measured | 30.7% | Only Qwen3 14B measured |
| Terminal-Bench Hard (Artificial Analysis run) | 3.8% | 5.3% | Qwen3 14B |
| LMArena Text: coding | 1231 | Not measured | Only Phi 4 measured |
| LMArena Text: software and IT | 1228 | Not measured | Only Phi 4 measured |
| Neither model has a result on the other 33 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Phi 4 | Qwen3 14B | Leader |
|---|---|---|---|
| AA Intelligence Index | 5.9 | 6.7 | Qwen3 14B |
| Epoch Capabilities Index | 130.5 | 138.2 | Qwen3 14B |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Phi 4 | Qwen3 14B | Leader |
|---|---|---|---|
| τ²-Bench Telecom | 0.0% | 34.5% | Qwen3 14B |
| BALROG | 11.6% | Not measured | 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 index03
| Benchmark | Phi 4 | Qwen3 14B | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 57.5% | 60.4% | Qwen3 14B |
| Humanity's Last Exam (Artificial Analysis run) | 3.8% | 4.5% | Qwen3 14B |
| GPQA Diamond (Epoch AI run) | 56.1% | 63.8% | Qwen3 14B |
| LMArena Text: hard prompts | 1220 | Not measured | Only Phi 4 measured |
| LMArena Text: expert | 1203 | Not measured | Only Phi 4 measured |
| LMArena Text: hard prompts in English | 1230 | Not measured | Only Phi 4 measured |
| Neither model has a result on the other 15 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Phi 4 | Qwen3 14B | Leader |
|---|---|---|---|
| AA Math Index | 18.0 | 58.0 | Qwen3 14B |
| AIME 2025 (Artificial Analysis run) | 18.0% | 58.0% | Qwen3 14B |
| OTIS Mock AIME | 13.8% | 66.4% | Qwen3 14B |
| MATH level 5 | 64.9% | Not measured | Only Phi 4 measured |
| LMArena Text: math | 1246 | Not measured | Only Phi 4 measured |
| LMArena Text: mathematical work | 1254 | Not measured | Only Phi 4 measured |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Phi 4 | Qwen3 14B | Leader |
|---|---|---|---|
| MMLU Pro (Artificial Analysis run) | 71.4% | 77.4% | Qwen3 14B |
| MMLU | 84.8% | Not measured | Only Phi 4 measured |
| LMArena Text: science | 1206 | Not measured | Only Phi 4 measured |
| LMArena Text: medicine and healthcare | 1202 | Not measured | Only Phi 4 measured |
| LMArena Text: legal and government | 1241 | Not measured | Only Phi 4 measured |
| LMArena Text: business and finance | 1199 | Not measured | Only Phi 4 measured |
| Neither model has a result on the other 11 boards here | |||
Web designNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Phi 4 | Qwen3 14B | Leader |
|---|---|---|---|
| Neither model has a result on 9 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Phi 4 | Qwen3 14B | Leader |
|---|---|---|---|
| IFBench | 23.5% | 40.5% | Qwen3 14B |
| Long context reasoning (AA-LCR) | 0.0% | 0.0% | Tie |
| DTBench | Not measured | 64.0% | Only Qwen3 14B measured |
| LMCA | Not measured | 18.2% | Only Qwen3 14B measured |
| Fiction.LiveBench | Not measured | 62.5% | Only Qwen3 14B measured |
| Creative writing (Lech Mazur) | 62.6% | Not measured | Only Phi 4 measured |
| LMArena Text | 1217 | Not measured | Only Phi 4 measured |
| LMArena Text: instruction following | 1201 | Not measured | Only Phi 4 measured |
| LMArena Text: creative writing | 1181 | Not measured | Only Phi 4 measured |
| LMArena Text: English | 1231 | Not measured | Only Phi 4 measured |
| LMArena Text: non-English | 1197 | Not measured | Only Phi 4 measured |
| LMArena Text: Chinese | 1212 | Not measured | Only Phi 4 measured |
| LMArena Text: German | 1222 | Not measured | Only Phi 4 measured |
| LMArena Text: Japanese | 1158 | Not measured | Only Phi 4 measured |
| LMArena Text: Korean | 1151 | Not measured | Only Phi 4 measured |
| LMArena Text: French | 1224 | Not measured | Only Phi 4 measured |
| LMArena Text: Spanish | 1234 | Not measured | Only Phi 4 measured |
| LMArena Text: Russian | 1209 | Not measured | Only Phi 4 measured |
| LMArena Text: multi-turn | 1206 | Not measured | Only Phi 4 measured |
| LMArena Text: longer queries | 1217 | Not measured | Only Phi 4 measured |
| LMArena Text: writing and language | 1198 | Not measured | Only Phi 4 measured |
| LMArena Text: entertainment and media | 1162 | Not measured | Only Phi 4 measured |
| Neither model has a result on the other 24 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Phi 4 | Qwen3 14B | 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 | Qwen3 14B |
|---|---|---|
| Maker | Microsoft | Qwen |
| Weights | Open weights | Open weights |
| Released | January 10, 2025 | April 28, 2025 |
| Context window | 16K | 41K |
| Input $/Mtok | $0.07 | $0.11 |
| Output $/Mtok | $0.14 | $0.23 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 0 | 0 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Phi 4 better than Qwen3 14B for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Phi 4 has the edge where output tokens cost 1.6x less, and where what matters is LiveCodeBench (Artificial Analysis run) or Terminal-Bench Hard (Artificial Analysis run), where it leads.
Which is cheaper, Phi 4 or Qwen3 14B?
Phi 4 costs $0.07 per million input tokens and $0.14 per million output tokens; Qwen3 14B costs $0.11 and $0.23. Cost per solved task is known for neither of them.
Which has the bigger context window, Phi 4 or Qwen3 14B?
Qwen3 14B, with 41K tokens against 16K.
How many benchmarks is the Phi 4 vs Qwen3 14B comparison based on?
14 benchmarks have a result for both models as of October 21, 2026. Phi 4's coding index rests on 0 independent benchmarks and Qwen3 14B's on 0. Where only one model has a result, the page says so instead of naming a winner.