Enter opens the highlighted result. Esc closes.

Kimi K2.5 vs Kimi K3

Kimi K3 leads Kimi K2.5 on coding, 62.3 to 38.6 on the Vibe Coding Index, and Kimi K2.5 costs 5.3x less per output token ($2.85 vs $15.00 per million). Updated October 22, 2026; 74 benchmarks have a result for both.

Replace Kimi K2.5 Replace Kimi K3
Kimi K2.5Moonshot AI
Coding
38.6
Overall
44.8
Output $/M
$2.85
1
Kimi K3Moonshot AI
Coding
62.3
Overall
66.5
Output $/M
$15.00
73
vs boards won of 74 shared

Kimi K2.5 or Kimi K3: which to pick

Pick Kimi K3 for coding: it scores 62.3 on the Vibe Coding Index against 38.6 for Kimi K2.5.

Pick Kimi K2.5 for output tokens 5.3x cheaper ($2.85 vs $15.00 per million) and a lower cost per solved task ($0.24 against $0.81).

Model AKimi K2.5Moonshot AI38.6/1003 independent benchmarks$0.24 per solved taskModel BKimi K3Moonshot AI62.3/1006 independent benchmarks$0.81 per solved task
Shared results
74
Kimi K2.5 only
25
Kimi K3 only
43
Like-for-like categories
8 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.

At a glance

Coding index
Kimi K2.5 38.6 from 3 benchmarks, Kimi K3 62.3 from 6 benchmarks.
Widest category
Coding, 25.2 points apart on average over 14 shared benchmarks.
Widest single gap
Vibe Code Bench (Vals AI): Kimi K2.5 17.5% against Kimi K3 85.0%, 67.4 points apart on a 0 to 100 scale.
Output price
Kimi K2.5 $2.85 against Kimi K3 $15.00 per million output tokens, 5.3x apart.
Context window
Kimi K2.5 takes 262K tokens, Kimi K3 1.05M.

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 K2.5BarsKimi K3Ties
Coding014
Intelligence02
Agentic work09
Reasoning010
Math02
Knowledge19
Web design07
General020
All shared boards173
CodingLike-for-likeSix of these boards feed the Vibe Coding Index014
Coding benchmarks for Kimi K2.5 and Kimi K3
BenchmarkKimi K2.5Kimi K3Leader
SWE-bench Verified70.8%Not measuredOnly Kimi K2.5 measured
SWE-bench Multilingual67.3%Not measuredOnly Kimi K2.5 measured
LiveBench codingNot measured59.3%Only Kimi K3 measured
LMArena WebDev14371655Kimi K3
AA Coding Index46.876.2Kimi K3
LiveCodeBench (Vals AI)83.9%87.2%Kimi K3
SWE-bench Verified (Vals AI)70.0%93.4%Kimi K3
Terminal-Bench 4 (Vals AI)Not measured17.2%Only Kimi K3 measured
Vibe Code Bench (Vals AI)17.5%85.0%Kimi K3
Code migration (Vals AI)7.0%16.1%Kimi K3
ProgramBench (Vals AI)Not measured2.0%Only Kimi K3 measured
IOI (Vals AI)Not measured48.9%Only Kimi K3 measured
Terminal-Bench 2.1 (Vals AI)42.0%80.9%Kimi K3
SWE-bench Multimodal (Vals AI)Not measured44.2%Only Kimi K3 measured
Vibe Code Bench 1-100 (Vals AI)Not measured18.2%Only Kimi K3 measured
CyberBench (Vals AI)Not measured75.2%Only Kimi K3 measured
SciCodeNot measured59.5%Only Kimi K3 measured
Terminal-Bench Hard (Artificial Analysis run)34.8%Not measuredOnly Kimi K2.5 measured
Terminal-Bench43.2%Not measuredOnly Kimi K2.5 measured
FrontierCodeNot measured44.2%Only Kimi K3 measured
FrontierSWENot measured25.9%Only Kimi K3 measured
WeirdML45.6%82.6%Kimi K3
ALE-Bench821.61524.5Kimi K3
LMArena Text: coding14831508Kimi K3
LMArena Text: software and IT14711499Kimi K3
LMArena WebDev: React14291670Kimi K3
LMArena WebDev: HTML14261644Kimi K3
LMArena WebDev: from an image14391579Kimi K3
Neither model has a result on the other 11 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for Kimi K2.5 and Kimi K3
BenchmarkKimi K2.5Kimi K3Leader
AA Intelligence Index23.543.6Kimi K3
Epoch Capabilities Index148.1157.9Kimi K3
Agentic workLike-for-likePublic boards, shown as published, kept out of the index09
Agentic work benchmarks for Kimi K2.5 and Kimi K3
BenchmarkKimi K2.5Kimi K3Leader
AA Agentic IndexNot measured50.0Only Kimi K3 measured
Design Arena: agents building web apps11361295Kimi K3
Design Arena: agents building full-stack apps11141315Kimi K3
Design Arena: agents building mobile apps11471269Kimi K3
Design Arena: agents building Android apps11181253Kimi K3
Design Arena: agents building Godot games12191223Kimi K3
Design Arena: agents building gamesNot measured1252Only Kimi K3 measured
Design Arena: HTML slidesNot measured1206Only Kimi K3 measured
Design Arena: PowerPoint slidesNot measured1247Only Kimi K3 measured
Design Arena: spreadsheetsNot measured1070Only Kimi K3 measured
LiveBench researchNot measured79.0%Only Kimi K3 measured
LiveBench browserNot measured16.0%Only Kimi K3 measured
LiveBench long agenticNot measured42.8%Only Kimi K3 measured
Finance Agent (Vals AI)35.8%53.1%Kimi K3
Tax Agent Bench (Vals AI)Not measured33.4%Only Kimi K3 measured
Legal Research Bench (Vals AI)15.9%46.1%Kimi K3
Harvey's Legal Agent Benchmark (Vals AI)Not measured12.9%Only Kimi K3 measured
Excel Modeling Benchmark (Vals AI)28.5%66.7%Kimi K3
BioMysteryBench (Vals AI)Not measured72.6%Only Kimi K3 measured
Terminal-Bench-Science (Vals AI)Not measured1.4%Only Kimi K3 measured
Time Horizon Index (Vals AI)Not measured10.5%Only Kimi K3 measured
τ²-Bench Telecom95.9%Not measuredOnly Kimi K2.5 measured
APEX-AgentsNot measured47.8%Only Kimi K3 measured
PostTrainBenchNot measured32.0%Only Kimi K3 measured
OSWorld63.3%Not measuredOnly Kimi K2.5 measured
Vending-Bench 2$1,198$5,165Kimi K3
Neither model has a result on the other 15 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index010
Reasoning benchmarks for Kimi K2.5 and Kimi K3
BenchmarkKimi K2.5Kimi K3Leader
GPQA Diamond84.1%92.9%Kimi K3
LiveBench reasoningNot measured90.7%Only Kimi K3 measured
MMMU Pro (Vals AI)84.3%88.2%Kimi K3
GPQA Diamond (Artificial Analysis run)87.9%93.5%Kimi K3
Humanity's Last Exam (Artificial Analysis run)30.7%46.9%Kimi K3
GPQA Diamond (Epoch AI run)Not measured93.1%Only Kimi K3 measured
ARC-AGI65.3%94.5%Kimi K3
ARC-AGI-211.8%60.4%Kimi K3
SimpleBench46.8%60.7%Kimi K3
Chess puzzlesNot measured37.0%Only Kimi K3 measured
Mystery game puzzlesNot measured26.0%Only Kimi K3 measured
LMArena Text: hard prompts14531495Kimi K3
LMArena Text: expert14661520Kimi K3
LMArena Text: hard prompts in English14601496Kimi K3
LMArena Vision: diagrams1282Not measuredOnly Kimi K2.5 measured
LMArena Vision: homework1297Not measuredOnly Kimi K2.5 measured
Kagi LLM Benchmark78.5%Not measuredOnly Kimi K2.5 measured
Neither model has a result on the other 5 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index02
Math benchmarks for Kimi K2.5 and Kimi K3
BenchmarkKimi K2.5Kimi K3Leader
AIME95.6%Not measuredOnly Kimi K2.5 measured
OTIS Mock AIMENot measured97.2%Only Kimi K3 measured
FrontierMath, tiers 1 to 3Not measured72.2%Only Kimi K3 measured
FrontierMath, tier 4Not measured39.0%Only Kimi K3 measured
ProofBenchNot measured87.0%Only Kimi K3 measured
LMArena Text: math14701494Kimi K3
LMArena Text: mathematical work14731505Kimi K3
Neither model has a result on the other 6 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index19
Knowledge benchmarks for Kimi K2.5 and Kimi K3
BenchmarkKimi K2.5Kimi K3Leader
MMLU Pro85.9%88.0%Kimi K3
LegalBench (Vals AI)Not measured86.2%Only Kimi K3 measured
CaseLaw (Vals AI)58.7%Not measuredOnly Kimi K2.5 measured
CorpFin (Vals AI)68.3%71.6%Kimi K3
TaxEval (Vals AI)74.2%75.7%Kimi K3
MortgageTax (Vals AI)66.5%66.3%Kimi K2.5
MedQA (Vals AI)94.4%Not measuredOnly Kimi K2.5 measured
MedCode (Vals AI)39.3%49.4%Kimi K3
SimpleQA Verified34.3%50.6%Kimi K3
LMArena Text: science14591508Kimi K3
LMArena Text: medicine and healthcare14481489Kimi K3
LMArena Text: legal and government14411503Kimi K3
LMArena Text: business and finance14271470Kimi K3
Neither model has a result on the other 5 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index07
Web design benchmarks for Kimi K2.5 and Kimi K3
BenchmarkKimi K2.5Kimi K3Leader
Design Arena: websites12481334Kimi K3
Design Arena: UI components12341341Kimi K3
Design Arena: game dev12061358Kimi K3
Design Arena: data viz12191351Kimi K3
Design Arena: 3D12041387Kimi K3
Design Arena: all code categories12361360Kimi K3
Design Arena: SVG11491302Kimi K3
Design Arena: ASCII art1166Not measuredOnly Kimi K2.5 measured
LiveBench designNot measured77.2%Only Kimi K3 measured
GeneralLike-for-likePublic boards, shown as published, kept out of the index020
General benchmarks for Kimi K2.5 and Kimi K3
BenchmarkKimi K2.5Kimi K3Leader
Vals IndexNot measured53.0%Only Kimi K3 measured
LiveBench data analysisNot measured78.7%Only Kimi K3 measured
Public Benefits Bench (Vals AI)Not measured68.2%Only Kimi K3 measured
SAGE (Vals AI)49.9%52.8%Kimi K3
MedScribe (Vals AI)76.4%88.0%Kimi K3
Vals Multimodal IndexNot measured73.4%Only Kimi K3 measured
IFBench70.2%Not measuredOnly Kimi K2.5 measured
Long context reasoning (AA-LCR)78.0%88.7%Kimi K3
DTBenchNot measured91.2%Only Kimi K3 measured
LMCANot measured52.7%Only Kimi K3 measured
Fiction.LiveBench86.1%Not measuredOnly Kimi K2.5 measured
CL-bench19.3%Not measuredOnly Kimi K2.5 measured
Surface Evolver BenchNot measured95.0%Only Kimi K3 measured
Furniture assemblyNot measured34.2%Only Kimi K3 measured
CL-bench Life13.2%Not measuredOnly Kimi K2.5 measured
LMArena Text14451476Kimi K3
LMArena Text: instruction following14311485Kimi K3
LMArena Text: creative writing14221457Kimi K3
LMArena Vision1269Not measuredOnly Kimi K2.5 measured
LMArena Text: English14531481Kimi K3
LMArena Text: non-English14331466Kimi K3
LMArena Text: Chinese14941529Kimi K3
LMArena Text: German14411490Kimi K3
LMArena Text: Japanese14231494Kimi K3
LMArena Text: Korean14121460Kimi K3
LMArena Text: French14541492Kimi K3
LMArena Text: Spanish14501472Kimi K3
LMArena Text: Russian14351485Kimi K3
LMArena Text: Polish14451458Kimi K3
LMArena Text: multi-turn14441489Kimi K3
LMArena Text: longer queries14451494Kimi K3
LMArena Text: writing and language14251471Kimi K3
LMArena Text: entertainment and media14171453Kimi K3
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 6 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 Kimi K3
BenchmarkKimi K2.5Kimi K3Leader
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$0.0020Fits in one request
Kimi K3$0.011Fits in one request

Kimi K2.5 costs less

Repository review

50K input + 3K output tokens

Kimi K2.5$0.035Fits in one request
Kimi K3$0.195Fits in one request

Kimi K2.5 costs less

Agent loop

200K input + 10K output tokens

Kimi K2.5$0.135Fits in one request
Kimi K3$0.750Fits in one request

Kimi K2.5 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 Kimi K3
SpecificationMoonshotAI Kimi K2.5MoonshotAI Kimi K3
MakerMoonshot AIMoonshot AI
WeightsOpen weightsOpen weights
ReleasedJanuary 27, 2026July 16, 2026
Context window262K1.05M
Input $/Mtok$0.532$3.00
Output $/Mtok$2.85$15.00
Coding index38.662.3
Independent measurements behind it36
Cost per solved task$0.24$0.81
SourceOpenRouter listingOpenRouter listing

Questions

Is Kimi K2.5 better than Kimi K3 for coding?

Pick Kimi K3 for coding: it scores 62.3 on the Vibe Coding Index against 38.6 for Kimi K2.5. Pick Kimi K2.5 for output tokens 5.3x cheaper ($2.85 vs $15.00 per million) and a lower cost per solved task ($0.24 against $0.81).

Which is cheaper, Kimi K2.5 or Kimi K3?

Kimi K2.5 costs $0.532 per million input tokens and $2.85 per million output tokens; Kimi K3 costs $3.00 and $15.00. Per solved SWE-bench Verified task, Kimi K2.5 costs $0.24 and Kimi K3 $0.81.

Which has the bigger context window, Kimi K2.5 or Kimi K3?

Kimi K3, with 1.05M tokens against 262K.

How many benchmarks is the Kimi K2.5 vs Kimi K3 comparison based on?

74 benchmarks have a result for both models as of October 22, 2026. Kimi K2.5's coding index rests on 3 independent benchmarks and Kimi K3's on 6. Where only one model has a result, the page says so instead of naming a winner.

Compare next

Embed this comparison on your site

Free under CC BY 4.0, while the source link stays visible. Preview the frame · Terms and other embeds

<iframe src="https://llmbenchmark.ai/embed/compare/kimi-k2-5-vs-kimi-k3" title="Kimi K2.5 vs Kimi K3, LLMBenchmark.ai" width="100%" height="360" style="border:0;max-width:100%" loading="lazy"></iframe>
<p style="font:13px/1.4 sans-serif;margin:6px 0 0">Source: <a href="https://llmbenchmark.ai/compare/kimi-k2-5-vs-kimi-k3">LLMBenchmark.ai, Kimi K2.5 vs Kimi K3</a></p>