Enter opens the highlighted result. Esc closes.

Phi 4 vs Qwen3 32B

Updated October 21, 2026. 37 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.

Replace Phi 4 Replace Qwen3 32B
Phi 4Microsoft
Coding
No data
Overall
5.9
Output $/M
$0.14
1
Qwen3 32BQwen
Coding
No data
Overall
7.3
Output $/M
$0.40
35
vs boards won of 37 shared, 1 tied

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 2.9x less, and where what matters is LiveCodeBench (Artificial Analysis run) or LMArena Text: coding, where it leads.

Model APhi 4MicrosoftNot benchmarkedNo coding benchmarkNo cost per solved taskModel BQwen3 32BQwenNot benchmarkedNo coding benchmarkNo cost per solved task
Shared results
37
Phi 4 only
8
Qwen3 32B only
5
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.

At a glance

Coding index
Phi 4 not benchmarked, Qwen3 32B not benchmarked.
Widest category
Math, 43.3 points apart on average over 5 shared benchmarks.
Widest single gap
AA Math Index: Phi 4 18.0 against Qwen3 32B 73.0, 55.0 points apart on a 0 to 100 scale.
Output price
Phi 4 $0.14 against Qwen3 32B $0.40 per million output tokens, 2.9x apart.
Context window
Phi 4 takes 16K tokens, Qwen3 32B 131K.

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.

Boards won, where both have a result
CategoryPhi 4BarsQwen3 32BTies
Coding13
Intelligence02
Agentic work01
Reasoning06
Math05
Knowledge05
General0131 tie
All shared boards1351 tie
CodingLike-for-likeSix of these boards feed the Vibe Coding Index13
Coding benchmarks for Phi 4 and Qwen3 32B
BenchmarkPhi 4Qwen3 32BLeader
Aider polyglotNot measured40.0%Only Qwen3 32B measured
LiveCodeBench (Artificial Analysis run)23.1%54.6%Qwen3 32B
SciCodeNot measured36.0%Only Qwen3 32B measured
Terminal-Bench Hard (Artificial Analysis run)3.8%3.0%Phi 4
LMArena Text: coding12311357Qwen3 32B
LMArena Text: software and IT12281366Qwen3 32B
Neither model has a result on the other 32 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for Phi 4 and Qwen3 32B
BenchmarkPhi 4Qwen3 32BLeader
AA Intelligence Index5.97.3Qwen3 32B
Epoch Capabilities Index130.5138.4Qwen3 32B
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
Agentic work benchmarks for Phi 4 and Qwen3 32B
BenchmarkPhi 4Qwen3 32BLeader
τ²-Bench Telecom0.0%29.8%Qwen3 32B
BALROG11.6%Not measuredOnly 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 index06
Reasoning benchmarks for Phi 4 and Qwen3 32B
BenchmarkPhi 4Qwen3 32BLeader
GPQA Diamond (Artificial Analysis run)57.5%66.8%Qwen3 32B
Humanity's Last Exam (Artificial Analysis run)3.8%7.4%Qwen3 32B
GPQA Diamond (Epoch AI run)56.1%65.7%Qwen3 32B
LMArena Text: hard prompts12201334Qwen3 32B
LMArena Text: expert12031362Qwen3 32B
LMArena Text: hard prompts in English12301350Qwen3 32B
Neither model has a result on the other 15 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index05
Math benchmarks for Phi 4 and Qwen3 32B
BenchmarkPhi 4Qwen3 32BLeader
AA Math Index18.073.0Qwen3 32B
AIME 2025 (Artificial Analysis run)18.0%73.0%Qwen3 32B
OTIS Mock AIME13.8%66.9%Qwen3 32B
MATH level 564.9%Not measuredOnly Phi 4 measured
LMArena Text: math12461399Qwen3 32B
LMArena Text: mathematical work12541420Qwen3 32B
Neither model has a result on the other 7 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index05
Knowledge benchmarks for Phi 4 and Qwen3 32B
BenchmarkPhi 4Qwen3 32BLeader
MMLU Pro (Artificial Analysis run)71.4%79.8%Qwen3 32B
MMLU84.8%Not measuredOnly Phi 4 measured
LMArena Text: science12061372Qwen3 32B
LMArena Text: medicine and healthcare12021368Qwen3 32B
LMArena Text: legal and government12411324Qwen3 32B
LMArena Text: business and finance11991318Qwen3 32B
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
Web design benchmarks for Phi 4 and Qwen3 32B
BenchmarkPhi 4Qwen3 32BLeader
Neither model has a result on 9 boards here
GeneralLike-for-likePublic boards, shown as published, kept out of the index013
General benchmarks for Phi 4 and Qwen3 32B
BenchmarkPhi 4Qwen3 32BLeader
IFBench23.5%36.3%Qwen3 32B
Long context reasoning (AA-LCR)0.0%0.0%Tie
DTBenchNot measured67.5%Only Qwen3 32B measured
LMCANot measured17.3%Only Qwen3 32B measured
Fiction.LiveBenchNot measured74.2%Only Qwen3 32B measured
Creative writing (Lech Mazur)62.6%Not measuredOnly Phi 4 measured
LMArena Text12171340Qwen3 32B
LMArena Text: instruction following12011305Qwen3 32B
LMArena Text: creative writing11811297Qwen3 32B
LMArena Text: English12311359Qwen3 32B
LMArena Text: non-English11971317Qwen3 32B
LMArena Text: Chinese12121357Qwen3 32B
LMArena Text: German12221341Qwen3 32B
LMArena Text: Japanese1158Not measuredOnly Phi 4 measured
LMArena Text: Korean1151Not measuredOnly Phi 4 measured
LMArena Text: French1224Not measuredOnly Phi 4 measured
LMArena Text: Spanish1234Not measuredOnly Phi 4 measured
LMArena Text: Russian12091311Qwen3 32B
LMArena Text: multi-turn12061331Qwen3 32B
LMArena Text: longer queries12171327Qwen3 32B
LMArena Text: writing and language11981298Qwen3 32B
LMArena Text: entertainment and media11621279Qwen3 32B
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
Image generation benchmarks for Phi 4 and Qwen3 32B
BenchmarkPhi 4Qwen3 32BLeader
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$0.0001Fits in one request
Qwen3 32B$0.0003Fits in one request

Phi 4 costs less

Repository review

50K input + 3K output tokens

Phi 4$0.0039Exceeds the context window
Qwen3 32B$0.0082Fits in one request

Phi 4 costs less

Agent loop

200K input + 10K output tokens

Phi 4$0.015Exceeds the context window
Qwen3 32B$0.032Exceeds the context window

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.

Specifications of Phi 4 and Qwen3 32B
SpecificationMicrosoft Phi 4Qwen3 32B
MakerMicrosoftQwen
WeightsOpen weightsOpen weights
ReleasedJanuary 10, 2025April 28, 2025
Context window16K131K
Input $/Mtok$0.07$0.14
Output $/Mtok$0.14$0.40
Coding indexNot benchmarkedNot benchmarked
Independent measurements behind it00
Cost per solved taskNot computableNot computable
SourceOpenRouter listingOpenRouter listing

Questions

Is Phi 4 better than Qwen3 32B 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 2.9x less, and where what matters is LiveCodeBench (Artificial Analysis run) or LMArena Text: coding, where it leads.

Which is cheaper, Phi 4 or Qwen3 32B?

Phi 4 costs $0.07 per million input tokens and $0.14 per million output tokens; Qwen3 32B costs $0.14 and $0.40. Cost per solved task is known for neither of them.

Which has the bigger context window, Phi 4 or Qwen3 32B?

Qwen3 32B, with 131K tokens against 16K.

How many benchmarks is the Phi 4 vs Qwen3 32B comparison based on?

37 benchmarks have a result for both models as of October 21, 2026. Phi 4's coding index rests on 0 independent benchmarks and Qwen3 32B's on 0. Where only one model has a result, the page says so instead of naming a winner.

Compare next