LLMBenchmark.ai

Enter opens the highlighted result. Esc closes.

Kimi K2.5 vs Qwen3 32B

Updated October 21, 2026. 32 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.

Replace Kimi K2.5 Replace Qwen3 32B
Kimi K2.5Moonshot AI
Coding
58.5
Overall
23.5
Output $/M
$2.50
32
Qwen3 32BQwen
Coding
40.0
Overall
7.3
Output $/M
$0.28
0
vs boards won of 32 shared

Decision reading

Pick Kimi K2.5 for coding: it scores 58.5 on the Vibe Coding Index against 40.0 for Qwen3 32B.

Pick Qwen3 32B when output tokens cost 8.9x less.

Model AKimi K2.5Moonshot AI58.5/1002 independent benchmarks$0.24 per solved taskModel BQwen3 32BQwen40.0/1001 independent benchmarkNo cost per solved task
Shared results
32
Kimi K2.5 only
66
Qwen3 32B 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.

At a glance

Coding index
Kimi K2.5 58.5 from 2 benchmarks, Qwen3 32B 40.0 from 1 benchmark.
Widest category
Agentic work, 66.1 points apart on average over 1 shared benchmark.
Widest single gap
Long context reasoning (AA-LCR): Kimi K2.5 78.0% against Qwen3 32B 0.0%, 78.0 points apart on a 0 to 100 scale.
Output price
Kimi K2.5 $2.50 against Qwen3 32B $0.28 per million output tokens, 8.9x apart.
Context window
Kimi K2.5 takes 262K 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
CategoryKimi K2.5BarsQwen3 32BTies
Coding30
Intelligence20
Agentic work10
Reasoning50
Math20
Knowledge40
General150
All shared boards320
CodingLike-for-likeFour of these boards feed the Vibe Coding Index30
Coding benchmarks for Kimi K2.5 and Qwen3 32B
BenchmarkKimi K2.5Qwen3 32BLeader
Aider polyglotNot measured40.0%Only Qwen3 32B measured
SWE-bench Verified70.8%Not measuredOnly Kimi K2.5 measured
SWE-bench Multilingual67.3%Not measuredOnly Kimi K2.5 measured
LMArena WebDev1437Not measuredOnly Kimi K2.5 measured
AA Coding Index46.8Not measuredOnly Kimi K2.5 measured
LiveCodeBench (Vals AI)83.9%Not measuredOnly Kimi K2.5 measured
SWE-bench Verified (Vals AI)70.0%Not measuredOnly Kimi K2.5 measured
Vibe Code Bench (Vals AI)17.5%Not measuredOnly Kimi K2.5 measured
Code migration (Vals AI)7.0%Not measuredOnly Kimi K2.5 measured
Terminal-Bench 2.1 (Vals AI)42.0%Not measuredOnly Kimi K2.5 measured
LiveCodeBench (Artificial Analysis run)Not measured54.6%Only Qwen3 32B measured
SciCodeNot measured36.0%Only Qwen3 32B measured
Terminal-Bench Hard (Artificial Analysis run)34.8%3.0%Kimi K2.5
Terminal-Bench43.2%Not measuredOnly Kimi K2.5 measured
WeirdML45.6%Not measuredOnly Kimi K2.5 measured
ALE-Bench821.6Not measuredOnly Kimi K2.5 measured
LMArena Text: coding14831357Kimi K2.5
LMArena Text: software and IT14711366Kimi K2.5
LMArena WebDev: React1429Not measuredOnly Kimi K2.5 measured
LMArena WebDev: HTML1426Not measuredOnly Kimi K2.5 measured
LMArena WebDev: from an image1439Not measuredOnly Kimi K2.5 measured
Neither model has a result on the other 17 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
Intelligence benchmarks for Kimi K2.5 and Qwen3 32B
BenchmarkKimi K2.5Qwen3 32BLeader
AA Intelligence Index23.57.3Kimi K2.5
Epoch Capabilities Index148.1138.4Kimi K2.5
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
Agentic work benchmarks for Kimi K2.5 and Qwen3 32B
BenchmarkKimi K2.5Qwen3 32BLeader
Design Arena: agents building web apps1136Not measuredOnly Kimi K2.5 measured
Design Arena: agents building full-stack apps1114Not measuredOnly Kimi K2.5 measured
Design Arena: agents building mobile apps1147Not measuredOnly Kimi K2.5 measured
Design Arena: agents building Android apps1124Not measuredOnly Kimi K2.5 measured
Design Arena: agents building Godot games1219Not measuredOnly Kimi K2.5 measured
Finance Agent (Vals AI)35.8%Not measuredOnly Kimi K2.5 measured
Legal Research Bench (Vals AI)15.9%Not measuredOnly Kimi K2.5 measured
Excel Modeling Benchmark (Vals AI)28.5%Not measuredOnly Kimi K2.5 measured
τ²-Bench Telecom95.9%29.8%Kimi K2.5
OSWorld63.3%Not measuredOnly Kimi K2.5 measured
Vending-Bench 2$1,198Not measuredOnly Kimi K2.5 measured
Neither model has a result on the other 28 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index50
Reasoning benchmarks for Kimi K2.5 and Qwen3 32B
BenchmarkKimi K2.5Qwen3 32BLeader
GPQA Diamond84.1%Not measuredOnly Kimi K2.5 measured
MMMU Pro (Vals AI)84.3%Not measuredOnly Kimi K2.5 measured
GPQA Diamond (Artificial Analysis run)87.9%66.8%Kimi K2.5
Humanity's Last Exam (Artificial Analysis run)30.7%7.4%Kimi K2.5
GPQA Diamond (Epoch AI run)Not measured65.7%Only Qwen3 32B measured
ARC-AGI65.3%Not measuredOnly Kimi K2.5 measured
ARC-AGI-211.8%Not measuredOnly Kimi K2.5 measured
SimpleBench46.8%Not measuredOnly Kimi K2.5 measured
LMArena Text: hard prompts14531334Kimi K2.5
LMArena Text: expert14661362Kimi K2.5
LMArena Text: hard prompts in English14601350Kimi K2.5
LMArena Vision: diagrams1282Not measuredOnly Kimi K2.5 measured
LMArena Vision: homework1297Not measuredOnly 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
Math benchmarks for Kimi K2.5 and Qwen3 32B
BenchmarkKimi K2.5Qwen3 32BLeader
AIME95.6%Not measuredOnly Kimi K2.5 measured
AA Math IndexNot measured73.0Only Qwen3 32B measured
AIME 2025 (Artificial Analysis run)Not measured73.0%Only Qwen3 32B measured
OTIS Mock AIMENot measured66.9%Only Qwen3 32B measured
LMArena Text: math14701399Kimi K2.5
LMArena Text: mathematical work14731420Kimi K2.5
Neither model has a result on the other 7 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index40
Knowledge benchmarks for Kimi K2.5 and Qwen3 32B
BenchmarkKimi K2.5Qwen3 32BLeader
MMLU Pro85.9%Not measuredOnly Kimi K2.5 measured
CaseLaw (Vals AI)58.7%Not measuredOnly Kimi K2.5 measured
CorpFin (Vals AI)68.3%Not measuredOnly Kimi K2.5 measured
TaxEval (Vals AI)74.2%Not measuredOnly Kimi K2.5 measured
MortgageTax (Vals AI)66.5%Not measuredOnly Kimi K2.5 measured
MedQA (Vals AI)94.4%Not measuredOnly Kimi K2.5 measured
MedCode (Vals AI)39.3%Not measuredOnly Kimi K2.5 measured
MMLU Pro (Artificial Analysis run)Not measured79.8%Only Qwen3 32B measured
SimpleQA Verified34.3%Not measuredOnly Kimi K2.5 measured
LMArena Text: science14591372Kimi K2.5
LMArena Text: medicine and healthcare14481368Kimi K2.5
LMArena Text: legal and government14411324Kimi K2.5
LMArena Text: business and finance14271318Kimi K2.5
Neither model has a result on the other 4 boards here
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
Web design benchmarks for Kimi K2.5 and Qwen3 32B
BenchmarkKimi K2.5Qwen3 32BLeader
Design Arena: websites1249Not measuredOnly Kimi K2.5 measured
Design Arena: UI components1236Not measuredOnly Kimi K2.5 measured
Design Arena: game dev1207Not measuredOnly Kimi K2.5 measured
Design Arena: data viz1219Not measuredOnly Kimi K2.5 measured
Design Arena: 3D1205Not measuredOnly Kimi K2.5 measured
Design Arena: all code categories1236Not measuredOnly Kimi K2.5 measured
Design Arena: SVG1149Not measuredOnly Kimi K2.5 measured
Design Arena: ASCII art1165Not measuredOnly 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 index150
General benchmarks for Kimi K2.5 and Qwen3 32B
BenchmarkKimi K2.5Qwen3 32BLeader
SAGE (Vals AI)49.9%Not measuredOnly Kimi K2.5 measured
MedScribe (Vals AI)76.4%Not measuredOnly Kimi K2.5 measured
IFBench70.2%36.3%Kimi K2.5
Long context reasoning (AA-LCR)78.0%0.0%Kimi K2.5
DTBenchNot measured67.5%Only Qwen3 32B measured
LMCANot measured17.3%Only Qwen3 32B measured
Fiction.LiveBench86.1%74.2%Kimi K2.5
CL-bench19.3%Not measuredOnly Kimi K2.5 measured
CL-bench Life13.2%Not measuredOnly Kimi K2.5 measured
LMArena Text14451340Kimi K2.5
LMArena Text: instruction following14311305Kimi K2.5
LMArena Text: creative writing14221297Kimi K2.5
LMArena Vision1269Not measuredOnly Kimi K2.5 measured
LMArena Text: English14531359Kimi K2.5
LMArena Text: non-English14331317Kimi K2.5
LMArena Text: Chinese14941357Kimi K2.5
LMArena Text: German14411341Kimi K2.5
LMArena Text: Japanese1423Not measuredOnly Kimi K2.5 measured
LMArena Text: Korean1412Not measuredOnly Kimi K2.5 measured
LMArena Text: French1454Not measuredOnly Kimi K2.5 measured
LMArena Text: Spanish1450Not measuredOnly Kimi K2.5 measured
LMArena Text: Russian14351311Kimi K2.5
LMArena Text: Polish1445Not measuredOnly Kimi K2.5 measured
LMArena Text: multi-turn14441331Kimi K2.5
LMArena Text: longer queries14451327Kimi K2.5
LMArena Text: writing and language14251298Kimi K2.5
LMArena Text: entertainment and media14171279Kimi K2.5
LMArena Vision: English1274Not measuredOnly Kimi K2.5 measured
LMArena Vision: Chinese1328Not measuredOnly Kimi K2.5 measured
LMArena Vision: reading text1276Not measuredOnly Kimi K2.5 measured
LMArena Vision: humor1284Not measuredOnly Kimi K2.5 measured
LMArena Vision: captioning1257Not measuredOnly Kimi K2.5 measured
LMArena Vision: recognizing things1285Not measuredOnly Kimi K2.5 measured
LMArena Vision: creative writing1269Not measuredOnly Kimi K2.5 measured
Neither model has a result on the other 12 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for Kimi K2.5 and Qwen3 32B
BenchmarkKimi K2.5Qwen3 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

Kimi K2.5$0.0018Fits in one request
Qwen3 32B$0.0002Fits in one request

Qwen3 32B costs less

Repository review

50K input + 3K output tokens

Kimi K2.5$0.033Fits in one request
Qwen3 32B$0.0048Fits in one request

Qwen3 32B costs less

Agent loop

200K input + 10K output tokens

Kimi K2.5$0.125Fits in one request
Qwen3 32B$0.019Exceeds the context window

Qwen3 32B 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 Kimi K2.5 and Qwen3 32B
SpecificationMoonshotAI Kimi K2.5Qwen3 32B
MakerMoonshot AIQwen
WeightsOpen weightsOpen weights
ReleasedJanuary 27, 2026April 28, 2025
Context window262K131K
Input $/Mtok$0.50$0.08
Output $/Mtok$2.50$0.28
Coding index58.540.0
Independent measurements behind it21
Cost per solved task$0.24Not computable
SourceOpenRouter listingOpenRouter listing

Questions

Is Kimi K2.5 better than Qwen3 32B for coding?

Pick Kimi K2.5 for coding: it scores 58.5 on the Vibe Coding Index against 40.0 for Qwen3 32B. Pick Qwen3 32B when output tokens cost 8.9x less.

Which is cheaper, Kimi K2.5 or Qwen3 32B?

Kimi K2.5 costs $0.50 per million input tokens and $2.50 per million output tokens; Qwen3 32B costs $0.08 and $0.28. Cost per solved task is known for Kimi K2.5 only.

Which has the bigger context window, Kimi K2.5 or Qwen3 32B?

Kimi K2.5, with 262K tokens against 131K.

How many benchmarks is the Kimi K2.5 vs Qwen3 32B comparison based on?

32 benchmarks have a result for both models as of October 21, 2026. Kimi K2.5's coding index rests on 2 independent benchmarks and Qwen3 32B's on 1. Where only one model has a result, the page says so instead of naming a winner.

Compare next