Enter opens the highlighted result. Esc closes.

Gemma 4 31B vs Kimi K3

Kimi K3 leads Gemma 4 31B on coding, 62.3 to 39.0 on the Vibe Coding Index, and Gemma 4 31B costs 37.5x less per output token ($0.40 vs $15.00 per million). Updated October 22, 2026; 54 benchmarks have a result for both.

Replace Gemma 4 31B Replace Kimi K3
Gemma 4 31BGoogle
Coding
39.0
Overall
42.1
Output $/M
$0.40
1
Kimi K3Moonshot AI
Coding
62.3
Overall
66.5
Output $/M
$15.00
53
vs boards won of 54 shared

Gemma 4 31B or Kimi K3: which to pick

Pick Kimi K3 for coding: it scores 62.3 on the Vibe Coding Index against 39.0 for Gemma 4 31B.

Pick Gemma 4 31B for output tokens 37.5x cheaper ($0.40 vs $15.00 per million) and work measured by SAGE (Vals AI), where it leads.

Model AGemma 4 31BGoogle39.0/1002 independent benchmarksNo cost per solved taskModel BKimi K3Moonshot AI62.3/1006 independent benchmarks$0.81 per solved task
Shared results
54
Gemma 4 31B only
18
Kimi K3 only
68
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
Gemma 4 31B 39.0 from 2 benchmarks, Kimi K3 62.3 from 6 benchmarks.
Widest category
Agentic work, 30.5 points apart on average over 4 shared benchmarks.
Widest single gap
Surface Evolver Bench: Gemma 4 31B 30.6% against Kimi K3 95.0%, 64.4 points apart on a 0 to 100 scale.
Output price
Gemma 4 31B $0.40 against Kimi K3 $15.00 per million output tokens, 37.5x apart.
Context window
Gemma 4 31B 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
CategoryGemma 4 31BBarsKimi K3Ties
Coding010
Intelligence02
Agentic work04
Reasoning07
Math03
Knowledge06
Web design01
General120
All shared boards153
CodingLike-for-likeSix of these boards feed the Vibe Coding Index010
Coding benchmarks for Gemma 4 31B and Kimi K3
BenchmarkGemma 4 31BKimi K3Leader
LiveBench coding49.0%59.3%Kimi K3
LMArena WebDev13671655Kimi K3
AA Coding Index43.476.2Kimi K3
LiveCodeBench (Vals AI)Not measured87.2%Only Kimi K3 measured
SWE-bench Verified (Vals AI)Not measured93.4%Only Kimi K3 measured
Terminal-Bench 4 (Vals AI)Not measured17.2%Only Kimi K3 measured
Vibe Code Bench (Vals AI)Not measured85.0%Only Kimi K3 measured
Code migration (Vals AI)Not measured16.1%Only Kimi K3 measured
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)Not measured80.9%Only Kimi K3 measured
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
SciCode45.5%59.5%Kimi K3
Terminal-Bench Hard (Artificial Analysis run)36.4%Not measuredOnly Gemma 4 31B measured
FrontierCodeNot measured44.2%Only Kimi K3 measured
FrontierSWENot measured25.9%Only Kimi K3 measured
WeirdML52.3%82.6%Kimi K3
ALE-Bench925.51524.5Kimi K3
LMArena Text: coding14591508Kimi K3
LMArena Text: software and IT14631499Kimi K3
LMArena WebDev: React13591670Kimi K3
LMArena WebDev: HTML13801644Kimi K3
LMArena WebDev: from an imageNot measured1579Only Kimi K3 measured
SciCode8.9%Not measuredOnly Gemma 4 31B measured
SWE-Bench Pro V2Not measured97.7%Only Kimi K3 measured
Neither model has a result on the other 16 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for Gemma 4 31B and Kimi K3
BenchmarkGemma 4 31BKimi K3Leader
AA Intelligence Index14.743.6Kimi K3
Epoch Capabilities Index143.0157.9Kimi K3
Agentic workLike-for-likePublic boards, shown as published, kept out of the index04
Agentic work benchmarks for Gemma 4 31B and Kimi K3
BenchmarkGemma 4 31BKimi K3Leader
AA Agentic Index4.250.0Kimi K3
Design Arena: agents building web appsNot measured1295Only Kimi K3 measured
Design Arena: agents building full-stack appsNot measured1315Only Kimi K3 measured
Design Arena: agents building mobile appsNot measured1268Only Kimi K3 measured
Design Arena: agents building Android appsNot measured1252Only Kimi K3 measured
Design Arena: agents building Godot gamesNot measured1223Only Kimi K3 measured
Design Arena: agents building gamesNot measured1252Only Kimi K3 measured
Design Arena: HTML slidesNot measured1196Only Kimi K3 measured
Design Arena: PowerPoint slidesNot measured1245Only Kimi K3 measured
Design Arena: spreadsheetsNot measured1096Only Kimi K3 measured
LiveBench research44.4%79.0%Kimi K3
LiveBench browser4.0%16.0%Kimi K3
LiveBench long agentic13.3%42.8%Kimi K3
Finance Agent (Vals AI)Not measured53.1%Only Kimi K3 measured
Tax Agent Bench (Vals AI)Not measured33.4%Only Kimi K3 measured
Legal Research Bench (Vals AI)Not measured46.1%Only Kimi K3 measured
Harvey's Legal Agent Benchmark (Vals AI)Not measured12.9%Only Kimi K3 measured
Excel Modeling Benchmark (Vals AI)Not measured66.7%Only Kimi K3 measured
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 Telecom65.5%Not measuredOnly Gemma 4 31B measured
APEX-AgentsNot measured47.8%Only Kimi K3 measured
PostTrainBenchNot measured32.0%Only Kimi K3 measured
Vending-Bench 2Not measured$5,165Only Kimi K3 measured
MCP AtlasNot measured82.3%Only Kimi K3 measured
Neither model has a result on the other 16 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index07
Reasoning benchmarks for Gemma 4 31B and Kimi K3
BenchmarkGemma 4 31BKimi K3Leader
GPQA DiamondNot measured92.9%Only Kimi K3 measured
LiveBench reasoning68.3%90.7%Kimi K3
MMMU Pro (Vals AI)Not measured88.2%Only Kimi K3 measured
GPQA Diamond (Artificial Analysis run)85.7%93.5%Kimi K3
Humanity's Last Exam (Artificial Analysis run)23.6%46.9%Kimi K3
GPQA Diamond (Epoch AI run)75.8%93.1%Kimi K3
ARC-AGINot measured94.5%Only Kimi K3 measured
ARC-AGI-2Not measured60.4%Only Kimi K3 measured
SimpleBenchNot measured60.7%Only Kimi K3 measured
Chess puzzlesNot measured37.0%Only Kimi K3 measured
Mystery game puzzlesNot measured26.0%Only Kimi K3 measured
LMArena Text: hard prompts14481495Kimi K3
LMArena Text: expert14641520Kimi K3
LMArena Text: hard prompts in English14561496Kimi K3
LMArena Vision: diagrams1283Not measuredOnly Gemma 4 31B measured
LMArena Vision: homework1310Not measuredOnly Gemma 4 31B measured
Kagi LLM Benchmark63.5%Not measuredOnly Gemma 4 31B measured
Humanity's Last Exam (Scale run)Not measured22.2%Only Kimi K3 measured
Neither model has a result on the other 5 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index03
Math benchmarks for Gemma 4 31B and Kimi K3
BenchmarkGemma 4 31BKimi K3Leader
OTIS Mock AIME73.3%97.2%Kimi K3
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: math14631494Kimi K3
LMArena Text: mathematical work14651505Kimi K3
MathArenaNot measured48.0%Only Kimi K3 measured
Neither model has a result on the other 7 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index06
Knowledge benchmarks for Gemma 4 31B and Kimi K3
BenchmarkGemma 4 31BKimi K3Leader
MMLU ProNot measured88.0%Only Kimi K3 measured
LegalBench (Vals AI)Not measured86.2%Only Kimi K3 measured
CaseLaw (Vals AI)52.6%Not measuredOnly Gemma 4 31B measured
CorpFin (Vals AI)Not measured71.6%Only Kimi K3 measured
TaxEval (Vals AI)Not measured75.7%Only Kimi K3 measured
MortgageTax (Vals AI)61.4%66.3%Kimi K3
MedCode (Vals AI)Not measured49.4%Only Kimi K3 measured
SimpleQA Verified10.4%50.6%Kimi K3
LMArena Text: science14571508Kimi K3
LMArena Text: medicine and healthcare14511489Kimi K3
LMArena Text: legal and government14381503Kimi K3
LMArena Text: business and finance14301470Kimi K3
FACTS Benchmark Suite45.3%Not measuredOnly Gemma 4 31B measured
Neither model has a result on the other 7 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index01
Web design benchmarks for Gemma 4 31B and Kimi K3
BenchmarkGemma 4 31BKimi K3Leader
Design Arena: websitesNot measured1335Only Kimi K3 measured
Design Arena: UI componentsNot measured1340Only Kimi K3 measured
Design Arena: game devNot measured1360Only Kimi K3 measured
Design Arena: data vizNot measured1352Only Kimi K3 measured
Design Arena: 3DNot measured1387Only Kimi K3 measured
Design Arena: all code categoriesNot measured1360Only Kimi K3 measured
Design Arena: SVGNot measured1301Only Kimi K3 measured
LiveBench design58.4%77.2%Kimi K3
Neither model has a result on the other 1 board here
GeneralLike-for-likePublic boards, shown as published, kept out of the index120
General benchmarks for Gemma 4 31B and Kimi K3
BenchmarkGemma 4 31BKimi K3Leader
Vals IndexNot measured53.0%Only Kimi K3 measured
LiveBench data analysis53.3%78.7%Kimi K3
Public Benefits Bench (Vals AI)Not measured68.2%Only Kimi K3 measured
SAGE (Vals AI)55.0%52.8%Gemma 4 31B
MedScribe (Vals AI)Not measured88.0%Only Kimi K3 measured
Vals Multimodal IndexNot measured73.4%Only Kimi K3 measured
IFBench75.6%Not measuredOnly Gemma 4 31B measured
Long context reasoning (AA-LCR)69.7%88.7%Kimi K3
DTBench82.7%91.2%Kimi K3
LMCA39.3%52.7%Kimi K3
Surface Evolver Bench30.6%95.0%Kimi K3
Furniture assemblyNot measured34.2%Only Kimi K3 measured
LMArena Text14431476Kimi K3
LMArena Text: instruction following14331485Kimi K3
LMArena Text: creative writing14171457Kimi K3
LMArena Vision1276Not measuredOnly Gemma 4 31B measured
LMArena Text: English14511481Kimi K3
LMArena Text: non-English14301466Kimi K3
LMArena Text: Chinese14751529Kimi K3
LMArena Text: GermanNot measured1490Only Kimi K3 measured
LMArena Text: JapaneseNot measured1494Only Kimi K3 measured
LMArena Text: KoreanNot measured1460Only Kimi K3 measured
LMArena Text: French14361492Kimi K3
LMArena Text: Spanish14421472Kimi K3
LMArena Text: Russian14601485Kimi K3
LMArena Text: Polish14331458Kimi K3
LMArena Text: multi-turn14521489Kimi K3
LMArena Text: longer queries14461494Kimi K3
LMArena Text: writing and language14321471Kimi K3
LMArena Text: entertainment and media14051453Kimi K3
LMArena Vision: English1271Not measuredOnly Gemma 4 31B measured
LMArena Vision: Chinese1326Not measuredOnly Gemma 4 31B measured
LMArena Vision: reading text1282Not measuredOnly Gemma 4 31B measured
LMArena Vision: humor1280Not measuredOnly Gemma 4 31B measured
LMArena Vision: captioning1218Not measuredOnly Gemma 4 31B measured
LMArena Vision: recognizing things1226Not measuredOnly Gemma 4 31B measured
LMArena Vision: creative writing1285Not measuredOnly Gemma 4 31B measured
LMArena Document1425Not measuredOnly Gemma 4 31B measured
EQ-Bench 411201339Kimi K3
Neither model has a result on the other 9 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for Gemma 4 31B and Kimi K3
BenchmarkGemma 4 31BKimi 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

Gemma 4 31B$0.0003Fits in one request
Kimi K3$0.011Fits in one request

Gemma 4 31B costs less

Repository review

50K input + 3K output tokens

Gemma 4 31B$0.0082Fits in one request
Kimi K3$0.195Fits in one request

Gemma 4 31B costs less

Agent loop

200K input + 10K output tokens

Gemma 4 31B$0.032Fits in one request
Kimi K3$0.750Fits in one request

Gemma 4 31B 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 4 31B and Kimi K3
SpecificationGoogle Gemma 4 31BMoonshotAI Kimi K3
MakerGoogleMoonshot AI
WeightsOpen weightsOpen weights
ReleasedApril 2, 2026July 16, 2026
Context window262K1.05M
Input $/Mtok$0.14$3.00
Output $/Mtok$0.40$15.00
Coding index39.062.3
Independent measurements behind it26
Cost per solved taskNot computable$0.81
SourceOpenRouter listingOpenRouter listing

Questions

Is Gemma 4 31B better than Kimi K3 for coding?

Pick Kimi K3 for coding: it scores 62.3 on the Vibe Coding Index against 39.0 for Gemma 4 31B. Pick Gemma 4 31B for output tokens 37.5x cheaper ($0.40 vs $15.00 per million) and work measured by SAGE (Vals AI), where it leads.

Which is cheaper, Gemma 4 31B or Kimi K3?

Gemma 4 31B costs $0.14 per million input tokens and $0.40 per million output tokens; Kimi K3 costs $3.00 and $15.00. Cost per solved task is known for Kimi K3 only.

Which has the bigger context window, Gemma 4 31B or Kimi K3?

Kimi K3, with 1.05M tokens against 262K.

How many benchmarks is the Gemma 4 31B vs Kimi K3 comparison based on?

54 benchmarks have a result for both models as of October 22, 2026. Gemma 4 31B's coding index rests on 2 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/gemma-4-31b-it-vs-kimi-k3" title="Gemma 4 31B 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/gemma-4-31b-it-vs-kimi-k3">LLMBenchmark.ai, Gemma 4 31B vs Kimi K3</a></p>