Enter opens the highlighted result. Esc closes.

Gemma 3 4B vs Kimi K2.5

Kimi K2.5 leads Gemma 3 4B on the Artificial Analysis Intelligence Index, 23.5 to 4.8, and Gemma 3 4B costs 38.0x less per output token ($0.075 vs $2.85 per million). Updated October 21, 2026; 31 benchmarks have a result for both.

Replace Gemma 3 4B Replace Kimi K2.5
Gemma 3 4BGoogle
Coding
No data
Overall
4.8
Output $/M
$0.075
0
Kimi K2.5Moonshot AI
Coding
38.6
Overall
23.5
Output $/M
$2.85
31
vs boards won of 31 shared

Gemma 3 4B or Kimi K2.5: which to pick

Only Kimi K2.5 has a coding score, 38.6 on the Vibe Coding Index; Gemma 3 4B has none, so the two cannot be ranked on coding.

Pick Gemma 3 4B for output tokens 38.0x cheaper ($0.075 vs $2.85 per million).

Model AGemma 3 4BGoogleNot benchmarkedNo coding benchmarkNo cost per solved taskModel BKimi K2.5Moonshot AI38.6/1003 independent benchmarks$0.24 per solved task
Shared results
31
Gemma 3 4B only
8
Kimi K2.5 only
67
Like-for-like categories
7 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
Gemma 3 4B not benchmarked, Kimi K2.5 38.6 from 3 benchmarks.
Widest category
Agentic work, 90.9 points apart on average over 1 shared benchmark.
Widest single gap
τ²-Bench Telecom: Gemma 3 4B 5.0% against Kimi K2.5 95.9%, 90.9 points apart on a 0 to 100 scale.
Output price
Gemma 3 4B $0.075 against Kimi K2.5 $2.85 per million output tokens, 38.0x apart.
Context window
Gemma 3 4B takes 131K tokens, Kimi K2.5 262K.

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
CategoryGemma 3 4BBarsKimi K2.5Ties
Coding04
Intelligence02
Agentic work01
Reasoning05
Math02
Knowledge04
General013
All shared boards031
CodingLike-for-likeSix of these boards feed the Vibe Coding Index04
Coding benchmarks for Gemma 3 4B and Kimi K2.5
BenchmarkGemma 3 4BKimi K2.5Leader
SWE-bench VerifiedNot measured70.8%Only Kimi K2.5 measured
SWE-bench MultilingualNot measured67.3%Only Kimi K2.5 measured
LMArena WebDevNot measured1437Only Kimi K2.5 measured
AA Coding Index2.746.8Kimi K2.5
LiveCodeBench (Vals AI)Not measured83.9%Only Kimi K2.5 measured
SWE-bench Verified (Vals AI)Not measured70.0%Only Kimi K2.5 measured
Vibe Code Bench (Vals AI)Not measured17.5%Only Kimi K2.5 measured
Code migration (Vals AI)Not measured7.0%Only Kimi K2.5 measured
Terminal-Bench 2.1 (Vals AI)Not measured42.0%Only Kimi K2.5 measured
LiveCodeBench (Artificial Analysis run)11.2%Not measuredOnly Gemma 3 4B measured
Terminal-Bench Hard (Artificial Analysis run)0.8%34.8%Kimi K2.5
Terminal-BenchNot measured43.2%Only Kimi K2.5 measured
WeirdMLNot measured45.6%Only Kimi K2.5 measured
ALE-BenchNot measured821.6Only Kimi K2.5 measured
LMArena Text: coding12301483Kimi K2.5
LMArena Text: software and IT12691471Kimi K2.5
LMArena WebDev: ReactNot measured1429Only Kimi K2.5 measured
LMArena WebDev: HTMLNot measured1426Only Kimi K2.5 measured
LMArena WebDev: from an imageNot measured1439Only Kimi K2.5 measured
Neither model has a result on the other 19 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for Gemma 3 4B and Kimi K2.5
BenchmarkGemma 3 4BKimi K2.5Leader
AA Intelligence Index4.823.5Kimi K2.5
Epoch Capabilities Index115.9148.1Kimi K2.5
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
Agentic work benchmarks for Gemma 3 4B and Kimi K2.5
BenchmarkGemma 3 4BKimi K2.5Leader
Design Arena: agents building web appsNot measured1136Only Kimi K2.5 measured
Design Arena: agents building full-stack appsNot measured1114Only Kimi K2.5 measured
Design Arena: agents building mobile appsNot measured1147Only Kimi K2.5 measured
Design Arena: agents building Android appsNot measured1118Only Kimi K2.5 measured
Design Arena: agents building Godot gamesNot measured1219Only Kimi K2.5 measured
Finance Agent (Vals AI)Not measured35.8%Only Kimi K2.5 measured
Legal Research Bench (Vals AI)Not measured15.9%Only Kimi K2.5 measured
Excel Modeling Benchmark (Vals AI)Not measured28.5%Only Kimi K2.5 measured
τ²-Bench Telecom5.0%95.9%Kimi K2.5
OSWorldNot measured63.3%Only Kimi K2.5 measured
Vending-Bench 2Not measured$1,198Only 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 index05
Reasoning benchmarks for Gemma 3 4B and Kimi K2.5
BenchmarkGemma 3 4BKimi K2.5Leader
GPQA DiamondNot measured84.1%Only Kimi K2.5 measured
MMMU Pro (Vals AI)Not measured84.3%Only Kimi K2.5 measured
GPQA Diamond (Artificial Analysis run)29.1%87.9%Kimi K2.5
Humanity's Last Exam (Artificial Analysis run)5.3%30.7%Kimi K2.5
GPQA Diamond (Epoch AI run)23.2%Not measuredOnly Gemma 3 4B measured
ARC-AGINot measured65.3%Only Kimi K2.5 measured
ARC-AGI-2Not measured11.8%Only Kimi K2.5 measured
SimpleBenchNot measured46.8%Only Kimi K2.5 measured
LMArena Text: hard prompts12531453Kimi K2.5
LMArena Text: expert12231466Kimi K2.5
LMArena Text: hard prompts in English12531460Kimi K2.5
LMArena Vision: diagramsNot measured1282Only Kimi K2.5 measured
LMArena Vision: homeworkNot measured1297Only 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 index02
Math benchmarks for Gemma 3 4B and Kimi K2.5
BenchmarkGemma 3 4BKimi K2.5Leader
AIMENot measured95.6%Only Kimi K2.5 measured
AA Math Index12.7Not measuredOnly Gemma 3 4B measured
AIME 2025 (Artificial Analysis run)12.7%Not measuredOnly Gemma 3 4B measured
OTIS Mock AIME7.5%Not measuredOnly Gemma 3 4B measured
LMArena Text: math12391470Kimi K2.5
LMArena Text: mathematical work12561473Kimi K2.5
Neither model has a result on the other 7 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
Knowledge benchmarks for Gemma 3 4B and Kimi K2.5
BenchmarkGemma 3 4BKimi K2.5Leader
MMLU ProNot measured85.9%Only Kimi K2.5 measured
CaseLaw (Vals AI)Not measured58.7%Only Kimi K2.5 measured
CorpFin (Vals AI)Not measured68.3%Only Kimi K2.5 measured
TaxEval (Vals AI)Not measured74.2%Only Kimi K2.5 measured
MortgageTax (Vals AI)Not measured66.5%Only Kimi K2.5 measured
MedQA (Vals AI)Not measured94.4%Only Kimi K2.5 measured
MedCode (Vals AI)Not measured39.3%Only Kimi K2.5 measured
MMLU Pro (Artificial Analysis run)41.7%Not measuredOnly Gemma 3 4B measured
SimpleQA VerifiedNot measured34.3%Only Kimi K2.5 measured
LMArena Text: science13061459Kimi K2.5
LMArena Text: medicine and healthcare13421448Kimi K2.5
LMArena Text: legal and government13271441Kimi K2.5
LMArena Text: business and finance13111427Kimi 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 Gemma 3 4B and Kimi K2.5
BenchmarkGemma 3 4BKimi K2.5Leader
Design Arena: websitesNot measured1248Only Kimi K2.5 measured
Design Arena: UI componentsNot measured1236Only Kimi K2.5 measured
Design Arena: game devNot measured1207Only Kimi K2.5 measured
Design Arena: data vizNot measured1219Only Kimi K2.5 measured
Design Arena: 3DNot measured1205Only Kimi K2.5 measured
Design Arena: all code categoriesNot measured1236Only Kimi K2.5 measured
Design Arena: SVGNot measured1149Only Kimi K2.5 measured
Design Arena: ASCII artNot measured1166Only 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 index013
General benchmarks for Gemma 3 4B and Kimi K2.5
BenchmarkGemma 3 4BKimi K2.5Leader
SAGE (Vals AI)Not measured49.9%Only Kimi K2.5 measured
MedScribe (Vals AI)Not measured76.4%Only Kimi K2.5 measured
IFBench28.3%70.2%Kimi K2.5
Long context reasoning (AA-LCR)6.7%78.0%Kimi K2.5
DTBench50.9%Not measuredOnly Gemma 3 4B measured
LMCA2.8%Not measuredOnly Gemma 3 4B measured
Fiction.LiveBenchNot measured86.1%Only Kimi K2.5 measured
CL-benchNot measured19.3%Only Kimi K2.5 measured
CL-bench LifeNot measured13.2%Only Kimi K2.5 measured
LMArena Text12911445Kimi K2.5
LMArena Text: instruction following12391431Kimi K2.5
LMArena Text: creative writing12701422Kimi K2.5
LMArena VisionNot measured1269Only Kimi K2.5 measured
LMArena Text: English13001453Kimi K2.5
LMArena Text: non-English12731433Kimi K2.5
LMArena Text: ChineseNot measured1494Only Kimi K2.5 measured
LMArena Text: German12811441Kimi K2.5
LMArena Text: JapaneseNot measured1423Only Kimi K2.5 measured
LMArena Text: KoreanNot measured1412Only Kimi K2.5 measured
LMArena Text: FrenchNot measured1454Only Kimi K2.5 measured
LMArena Text: SpanishNot measured1450Only Kimi K2.5 measured
LMArena Text: Russian12941435Kimi K2.5
LMArena Text: PolishNot measured1445Only Kimi K2.5 measured
LMArena Text: multi-turn12551444Kimi K2.5
LMArena Text: longer queries12731445Kimi K2.5
LMArena Text: writing and language12881425Kimi K2.5
LMArena Text: entertainment and media12331417Kimi K2.5
LMArena Vision: EnglishNot measured1274Only Kimi K2.5 measured
LMArena Vision: ChineseNot measured1328Only Kimi K2.5 measured
LMArena Vision: reading textNot measured1276Only Kimi K2.5 measured
LMArena Vision: humorNot measured1284Only Kimi K2.5 measured
LMArena Vision: captioningNot measured1257Only Kimi K2.5 measured
LMArena Vision: recognizing thingsNot measured1285Only Kimi K2.5 measured
LMArena Vision: creative writingNot measured1269Only 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 Gemma 3 4B and Kimi K2.5
BenchmarkGemma 3 4BKimi K2.5Leader
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

Gemma 3 4B$0.0001Fits in one request
Kimi K2.5$0.0020Fits in one request

Gemma 3 4B costs less

Repository review

50K input + 3K output tokens

Gemma 3 4B$0.0025Fits in one request
Kimi K2.5$0.035Fits in one request

Gemma 3 4B costs less

Agent loop

200K input + 10K output tokens

Gemma 3 4B$0.0097Exceeds the context window
Kimi K2.5$0.135Fits in one request

Gemma 3 4B 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 Gemma 3 4B and Kimi K2.5
SpecificationGoogle Gemma 3 4BMoonshotAI Kimi K2.5
MakerGoogleMoonshot AI
WeightsOpen weightsOpen weights
ReleasedMarch 13, 2025January 27, 2026
Context window131K262K
Input $/Mtok$0.045$0.532
Output $/Mtok$0.075$2.85
Coding indexNot benchmarked38.6
Independent measurements behind it03
Cost per solved taskNot computable$0.24
SourceOpenRouter listingOpenRouter listing

Questions

Is Gemma 3 4B better than Kimi K2.5 for coding?

Only Kimi K2.5 has a coding score, 38.6 on the Vibe Coding Index; Gemma 3 4B has none, so the two cannot be ranked on coding. Pick Gemma 3 4B for output tokens 38.0x cheaper ($0.075 vs $2.85 per million).

Which is cheaper, Gemma 3 4B or Kimi K2.5?

Gemma 3 4B costs $0.045 per million input tokens and $0.075 per million output tokens; Kimi K2.5 costs $0.532 and $2.85. Cost per solved task is known for Kimi K2.5 only.

Which has the bigger context window, Gemma 3 4B or Kimi K2.5?

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

How many benchmarks is the Gemma 3 4B vs Kimi K2.5 comparison based on?

31 benchmarks have a result for both models as of October 21, 2026. Gemma 3 4B's coding index rests on 0 independent benchmarks and Kimi K2.5's on 3. Where only one model has a result, the page says so instead of naming a winner.

Compare next