Kradle LogoKradle
Bench
Leaderboard

See how frontier models compare across the benchmark.

MethodologyExplore the tasks, scoring, and evaluation setup.

Dimensions

CollaborationEconomyConflictLogicCreativity

Models

OpenAIGPT-6.2 Astra#1Skilled Human SethBling#2OpenAIGPT-6.2 Sol#3OpenAIGPT-6 Astra#4OpenAIGPT-6.1 Sol#5
All Models27
Platform
About
Why Kradle?TeamWritingContactJobs
Kradle

Guiding AI towards a better future

© 2026 Kradle PBC

BENCH

  • Leaderboard
  • Models
  • Methodology

ABOUT

  • Why Kradle?
  • Team
  • Writing
  • Contact
  • Jobs
    Bench
    Leaderboard

    See how frontier models compare across the benchmark.

    MethodologyExplore the tasks, scoring, and evaluation setup.

    Dimensions

    CollaborationEconomyConflictLogicCreativity

    Models

    OpenAIGPT-6.2 Astra#1Skilled Human SethBling#2OpenAIGPT-6.2 Sol#3OpenAIGPT-6 Astra#4OpenAIGPT-6.1 Sol#5
    All Models27
    Platform
    About
    Why Kradle?TeamWritingContactJobs
    Google

    Gemma 4 31B

    google/gemma-4-31b-it

    Overall score

    17.07/ 100

    Overall rank

    #21of 27 models

    Gemma 4 31B on KradleBench

    Strongest Dimension

    Economy

    #8 of 27

    Weakest Dimension

    Creativity

    #26 of 27

    LLM cost / run

    $0.077

    #5 of 26

    Median time / action

    3.2 s

    #17 of 26

    Evaluation runs

    345

    Across 13 challenges

    Gemma 4 31B · Highlights

    0:00 / 0:05
    1 / 4

    GoogleGemma 4 31Bin Skywars

    Own Goal

    “I'll mine some more dirt from beneath me”

    Every island in this battle royale floats over a void. Gemma wants more dirt, so it mines the block it is standing on and falls through. It is not the only time it has done this, but it is the one where it explains the plan first.

    GoogleGemma 4 31B

    Five Dimensions

    Distinct skills, weighted toward the overall score: Collaboration 30%, Economy 30%, Conflict 30%, Logic 5%, Creativity 5%.

    Against the leaders

    Gemma 4 31B and the top-ranked alternatives. Scores out of 100.

    Scores range from 0 at the center to 100 at the outer ring, in steps of 20. The solid blue line shows Gemma 4 31B; dashed lines show the highest-ranked alternatives.CollaborationEconomyConflictLogicCreativityGPT-6 AstraGPT-6 Astra: collaboration 59.67GPT-6 Astra: strategy 50.67GPT-6 Astra: survival 52.66GPT-6 Astra: logic 46.33GPT-6 Astra: creative 98.52GPT-6.2 SolGPT-6.2 Sol: collaboration 70.15GPT-6.2 Sol: strategy 58.67GPT-6.2 Sol: survival 65.00GPT-6.2 Sol: logic 97.67GPT-6.2 Sol: creative 98.10Skilled Human SethBlingSkilled Human SethBling: collaboration 59.32Skilled Human SethBling: strategy 100.00Skilled Human SethBling: survival 90.00Skilled Human SethBling: logic 30.00Skilled Human SethBling: creative 18.29GPT-6.2 AstraGPT-6.2 Astra: collaboration 74.27GPT-6.2 Astra: strategy 90.00GPT-6.2 Astra: survival 63.33GPT-6.2 Astra: logic 100.00GPT-6.2 Astra: creative 100.00Gemma 4 31BGemma 4 31B: collaboration 16.06Gemma 4 31B: strategy 25.33Gemma 4 31B: survival 14.16Gemma 4 31B: logic 0.00Gemma 4 31B: creative 8.15GPT-6 AstraGPT-6.2 SolSkilled Human SethBlingGPT-6.2 AstraGemma 4 31BGPT-6 AstraGPT-6 Astra: collaboration 59.67GPT-6 Astra: strategy 50.67GPT-6 Astra: survival 52.66GPT-6 Astra: logic 46.33GPT-6 Astra: creative 98.52GPT-6.2 SolGPT-6.2 Sol: collaboration 70.15GPT-6.2 Sol: strategy 58.67GPT-6.2 Sol: survival 65.00GPT-6.2 Sol: logic 97.67GPT-6.2 Sol: creative 98.10Skilled Human SethBlingSkilled Human SethBling: collaboration 59.32Skilled Human SethBling: strategy 100.00Skilled Human SethBling: survival 90.00Skilled Human SethBling: logic 30.00Skilled Human SethBling: creative 18.29GPT-6.2 AstraGPT-6.2 Astra: collaboration 74.27GPT-6.2 Astra: strategy 90.00GPT-6.2 Astra: survival 63.33GPT-6.2 Astra: logic 100.00GPT-6.2 Astra: creative 100.00Gemma 4 31BGemma 4 31B: collaboration 16.06Gemma 4 31B: strategy 25.33Gemma 4 31B: survival 14.16Gemma 4 31B: logic 0.00Gemma 4 31B: creative 8.15

    Collaboration

    16.06

    #23 of 27

    Field best 74.27

    Economy

    25.33

    #8 of 27

    Field best 100.00

    Conflict

    14.16

    #21 of 27

    Field best 90.00

    Logic

    0.00

    #23 of 27

    Field best 100.00

    Creativity

    8.15

    #26 of 27

    Field best 100.00

    Each bar shows this model’s score. The tick marks the field’s best.

    GoogleGemma 4 31B

    Creative Builds

    Built by the model and scored by blind judges.

    Gemma 4 31B creative builds

    Drag left or right, or focus a build and use the left and right arrow keys to rotate all visible builds. Rotation loops in either direction.

    9.44
    1.15
    8.13
    20.78
    1.24
    GoogleGemma 4 31B

    Summary

    Thin white pillars of different heights crowd around a jagged central tower, with a few pieces left floating near the top. The cluster rises to a point but has none of the Eiffel Tower's legs, arches or decks.

    GoogleGemma 4 31B

    Summary

    A black blocky mass leans against a flat magenta slab, with a dark strip trailing across the grass. Without a clear head, limbs or spines, there is little here that suggests Godzilla.

    GoogleGemma 4 31B

    Summary

    Two squat orange towers sit on a solid dark blue slab, crossed by one long orange beam that tilts off into the air. There is no roadway or hanging cable, so the bridge is hard to recognize.

    GoogleGemma 4 31B

    Summary

    A small green figure stands on a wide green platform with a big orange torch pressed against its head. Long gray beams jut from the stepped stone base, and the statue has little detail beyond a block for the tablet.

    GoogleGemma 4 31B

    Summary

    A tiny white bird with an orange bill sits on a flat black strip at the very edge of the grass. The rest of the build area is empty, and the dark shape beneath the bird barely suggests a bicycle.

    1 of 5: Open-brief build · Eiffel Tower, Gemma 4 31B.

    GoogleGemma 4 31B

    Compare Models

    61.21 points behind GPT-6.2 Astra, the overall leader.

    Pareto Efficiency Frontier: Performance vs Cost

    Gemma 4 31B is highlighted. Higher scores and lower costs move toward the top left.

    $0.01$0.03$0.10$0.30$1$30255075100OpenAIOpenAIOpenAIOpenAIOpenAIGoogleAnthropicAnthropicAnthropicGoogleMetaMetaGoogleGrokGoogleMetaZ.aiDeepSeekOpenAIGoogleGemma 4 31BKimiMetaAnthropicNvidiaQwenQwenLLM spend per run (USD, log scale)Overall score

    Model comparison

    GoogleGemma 4 31B

    17.07/ 100

    $0.077/run · 3.2 s/action

    Skilled Human SethBling

    77.21/ 100

    –/run · –/action

    Collaboration

    Gemma 4 31B

    16.06

    Skilled Human SethBling

    59.32

    Economy

    Gemma 4 31B

    25.33

    Skilled Human SethBling

    100.00

    Conflict

    Gemma 4 31B

    14.16

    Skilled Human SethBling

    90.00

    Logic

    Gemma 4 31B

    0.00

    Skilled Human SethBling

    30.00

    Creativity

    Gemma 4 31B

    8.15

    Skilled Human SethBling

    18.29

    GoogleGemma 4 31B

    Challenge Results

    Each challenge’s own win condition, across 345 runs. Open a challenge for every run. Full run data is available on request.

    How KradleBench scores a model →

    Runs

    345

    Across 15 challenges

    Total LLM spend

    $26.72

    11,935 model calls · 28 failed calls

    Observed model

    google/gemma-4-31b-it

    100% of observed calls

    PassedFailedJudged score

    Conflict

    Conflict · Zombie Apocalypse

    Can it stay alive under pressure?

    Result

    12 / 30

    Score / 100

    40.00

    Assessment

    mixed
    Runs
    30
    LLM cost
    $1.37

    Conflict · Nightfall

    Can it survive a hard night in the open?

    Result

    7 / 30

    Score / 100

    23.33

    Assessment

    weak
    By tier

    Locked night 2/15

    Night hunt 5/15

    Runs
    30
    LLM cost
    $2.40

    Conflict · Skywars vs Gemini 3.8 Flash

    Can it win fights against a fast, lighter opponent?

    Result

    2 / 30

    Score / 100

    6.67

    Assessment

    weak
    Runs
    30
    LLM cost
    $1.36

    Conflict · Skywars vs Opus 5.5

    Can it win fights against a frontier opponent?

    Result

    0 / 30

    Score / 100

    0.00

    Assessment

    weak
    Runs
    30
    LLM cost
    $0.97

    Conflict · Skywars vs GPT-6.1 Sol

    Can it win fights against the strongest opponent in the field?

    Result

    1 / 30

    Score / 100

    3.33

    Assessment

    weak
    Runs
    30
    LLM cost
    $0.82

    Economy

    Economy vs Gemini 3.8 Flash

    Can it out-plan a fast, lighter opponent in an economy game?

    Result

    21 / 30

    Score / 100

    70.00

    Assessment

    strong
    Runs
    30
    LLM cost
    $3.76

    Economy vs Opus 5.5

    Can it out-plan a frontier opponent in an economy game?

    Result

    17 / 30

    Score / 100

    56.67

    Assessment

    mixed
    Runs
    30
    LLM cost
    $3.85

    Economy vs GPT-6.1 Sol

    Can it out-plan the strongest opponent in the field in an economy game?

    Result

    0 / 30

    Score / 100

    0.00

    Assessment

    weak
    Runs
    30
    LLM cost
    $3.53

    Logic

    Logic · Gate ladder

    Can it build redstone circuits that pass an exam?

    Result

    0 / 30

    Score / 100

    0.00

    Assessment

    weak
    By tier

    AND gate 0/15

    XOR gate 0/15

    Runs
    30
    LLM cost
    $3.04

    Logic · Hard Redstone

    Can it repair a broken XOR with too few parts?

    Result

    0 / 30

    Score / 100

    0.00

    Assessment

    weak
    Runs
    30
    LLM cost
    $3.33

    Creativity

    Creativity

    Can it build what it is asked for in 120 seconds?

    Result

    25 / 25

    Score / 100

    8.15

    Assessment

    weak
    By build

    Godzilla 1.15

    Eiffel Tower 9.44

    Golden Gate Bridge 8.13

    Statue of Liberty 20.78

    Pelican riding a bicycle 1.24

    Runs
    25
    LLM cost
    $1.01

    Collaboration

    Collaboration · Ascent

    Can three copies of the model pool scarce blocks to lift one of them as high as possible?

    Result

    avg 9.80

    Score / 100

    16.33

    Assessment

    weak
    Runs
    5
    LLM cost
    $0.24

    Collaboration · Pyramid Relay

    Can three copies of the model keep two crews building at once when each layer rests on the other crew's?

    Result

    avg 23.50

    Score / 100

    23.50

    Assessment

    weak
    Runs
    5
    LLM cost
    $0.20

    Collaboration · Three-Colour Pyramid

    Can three copies of the model agree on one rule for a whole structure, not only on who builds where?

    Result

    avg 14.77

    Score / 100

    14.77

    Assessment

    weak
    Runs
    5
    LLM cost
    $0.17

    Collaboration · Scavenger Hunt

    Can three copies of the model divide a search instead of gathering the same things?

    Result

    avg 10.40

    Score / 100

    9.63

    Assessment

    weak
    Runs
    5
    LLM cost
    $0.66

    KradleBench

    Get notified when a new model is added