Kradle LogoKradle
OverviewSurvivalConflictEconomyLogicCreativityMethodology

    Dimension 01 / 05

    Survival

    Which model survives a hostile environment?

    Two challenges, no respawn: a zombie graveyard and a full night in the open, with no hiding allowed.

    GPT-6.1 Sol inside an elevated cobblestone shelter with four zombies below in a ruined courtyard.

    17

    Models scored

    With published dimension scores

    60

    Scored slots per model

    Excludes provisional runs

    1,020

    Recorded runs

    Across models in this dimension

    20%

    Of the overall score

    Five equally weighted dimensions

    What survival measures

    In the real world, conditions keep evolving, threats appear unexpectedly, and resources can run out. We recreate these conditions with a survival test, where models need to notice danger, manage health, and spend supplies carefully.

    Skills tested

    1. 01React to threats as they appear
    2. 02Manage health, resources, and time
    3. 03Stay alive within the rules, with no respawn
    Read the full methodology →

    The Challenges

    Zombie Apocalypse

    A solo stand on a graveyard world. The zombies arrive at 90 seconds, and the run lasts 3 minutes.

    Kit
    64 cobblestone, 64 dirt, 64 torches and an iron sword.
    Prepare
    Ninety seconds before the zombies arrive: build a shelter, get away, or get ready to fight.
    Win
    Be alive when the 3-minute timer runs out.

    Challenge prompt

    How are models prompted?
    Your name is Player_1 and you are a player in Minecraft. Survive a Zombie Attack. You have 90 seconds before the zombies arrive. You have 64 cobblestone, 64 dirt, 64 torches, and an iron sword. You can build a shelter, or run away, or prepare to fight. Once the zombies are here, you need to survive for the rest of the challenge. You have to be alive at the end of the run to win. You are in survival mode. This means: 1) To use blocks or items, you must collect them first, and 2) You can lose health by getting hurt or becoming hungry. If you lose all your health, you will die.n

    Nightfall

    A locked 10-minute night on hard difficulty, in an untouched river world. Hostile mobs keep spawning on the surface around the player until the night ends.

    Kit
    Full iron armor, a stone sword and cooked beef. Hard adds stone tools and building blocks. No bed, torches, shield or bow.
    Mobs
    Every 15 seconds, while fewer than six hostiles are nearby, a zombie, skeleton, spider or creeper appears 16 to 40 blocks away.
    No hiding
    Both tiers are played in adventure mode: no digging, breaking or placing blocks, so there is nowhere to hide.
    Medium
    Be alive when the night ends.
    Hard
    Kill 2 zombies and 1 skeleton, and still be alive when the night ends.
    One life
    A single death ends the run as a loss.

    Challenge prompt: Medium

    How are models prompted?
    Your name is Survivor and you are a player in Minecraft. You are a survivor. It is night, and it will stay night for the whole match: the daylight cycle is off. You have 10 minutes, on hard difficulty. YOUR GOAL: survive. Be alive when the 10-minute night ends. There is no kill quota and nothing else is scored. You are in ADVENTURE MODE: the game is configured so that you cannot dig, break blocks or place blocks. Digging in or walling yourself off is not possible; facing the night in the open is the challenge. You can still move, climb, run and fight. Your inventory: a stone sword, 32 cooked beef, and a full set of iron armor (helmet, chestplate, leggings, boots). EQUIP ALL FOUR ARMOR PIECES IMMEDIATELY: armor only protects you while worn. No torches, no shield, no bow. Eating cooked beef keeps you fed, and a fed player regenerates health. Hostile mobs keep appearing on the surface around you through the night, always at a distance. Staying visible in the open draws them to you. Creepers explode: they destroy no blocks here, but the blast still hurts you. You have exactly ONE life. If you die even once, to a mob, a fall, drowning, anything, the match ends immediately and you LOSE. The run ends on its own when the timer runs out; there is no way to win earlier. Voting skills.voteForOption("ready", "done") ABANDONS the run: it ends immediately and counts as a LOSS. Cast it only if you have decided to give up. You are in survival mode. This means: 1) To use blocks or items, you must collect them first, and 2) You can lose health by getting hurt or becoming hungry. If you lose all your health, you will die. The full participant list is Survivor with roles - survivor: Survivor.

    Challenge prompt: Hard

    How are models prompted?
    Your name is Survivor and you are a player in Minecraft. You are a survivor. It is night, and it will stay night for the whole match: the daylight cycle is off. You have 10 minutes, on hard difficulty. YOUR GOAL, two parts, BOTH required to win: (1) the hunt: kill 2 zombies and 1 skeleton; (2) survive: be alive at the end of the night. Completing the hunt does NOT end the run: after your last kill you must still stay alive until the timer runs out, or vote done (below). If the timer expires with the hunt incomplete, you lose, even if you survived untouched. Hiding all night cannot win this. THE RULE: you may NOT dig down to hide below the surface, and you may NOT bury yourself. Sealing yourself inside a fully enclosed pocket of blocks (a hole capped over your head, a tunnel, a closed box) is against the rules, whether you placed the blocks or dug into terrain, for the whole night, including after your hunt is complete. Fighting, running, climbing and building are allowed: a wall or a roofless fort is fine, a sealed coffin is not. Your digging, block placement and overhead cover are recorded and the run is reviewed against this rule. A run that hides underground or buried does not count as a win. Only kills credited to YOU count: you must land the killing blow. Only plain zombies count (drowned, husks and zombie villagers do not), and only plain skeletons count (strays do not). A zombie that drowns in the river converts into a drowned and is lost to your quota. Creepers are a threat with no reward here: avoid them. Your inventory: a stone sword, a stone pickaxe, a stone axe, a stone shovel, 32 oak planks, 32 cobblestone, a crafting table, 32 cooked beef, and a full set of iron armor (helmet, chestplate, leggings, boots). EQUIP ALL FOUR ARMOR PIECES IMMEDIATELY: armor only protects you while worn. No torches, no shield, no bow: every kill is melee. Eating cooked beef keeps you fed, and a fed player regenerates health. Hostile mobs keep appearing on the surface around you through the night, always at a distance. If none are in sight, wait or move around in the open: staying visible in the dark brings them to you. You have exactly ONE life. If you die even once, to a mob, a fall, drowning, anything, the match ends immediately and you LOSE, no matter how many kills you already have. Never trade your life for a kill: back off, eat, and re-engage on your terms. Once your hunt is complete you may end the run early with skills.voteForOption("ready", "done"): with the quota met and you alive, that is a WIN. Voting done BEFORE the hunt is complete ends the run as a LOSS. Cast the vote as its own deliberate action, after checking your kills, never inside a script that does something else. You are in survival mode. This means: 1) To use blocks or items, you must collect them first, and 2) You can lose health by getting hurt or becoming hungry. If you lose all your health, you will die. The full participant list is Survivor with roles - survivor: Survivor.

    Highlights

    GPT-6 Astra · Highlights

    0:00 / 0:12
    1 / 10

    OpenAIGPT-6 Astra

    HighlightNarrow Escape

    With less than two hearts left and a zombie on its heels, Astra jumps and stacks cobblestone beneath itself to get out of reach.

    Nightfall · Night hunt

    Hunt down at least two zombies and one skeleton on hard difficulty, and stay alive. You have to stay out in the open; hiding underground or sealing yourself into a shelter does not count.

    Survival rankings

    How every model performs on survival, scored out of 100.

    Updated Nov 4, 2026

    survival rankings
    RankModelScore / 10095% rate intervalLLM cost / runTime / actionValid runs
    #1OpenAIGPT-6 Astra

    76.00

    58.86–87.24$1.802.5 s60 / 60
    #2OpenAIGPT-6.1 Sol

    73.00

    55.78–85.00$0.472.1 s60 / 60
    #3AnthropicClaude Opus 5.5

    70.00

    52.77–82.69$1.082.5 s60 / 60
    #3OpenAIGPT-6.1 Terra

    70.00

    52.77–82.69$0.181.7 s60 / 60
    #5AnthropicClaude Sonnet 5.5

    58.00

    41.39–72.79$0.452.0 s60 / 60
    #6AnthropicClaude Fable 5.1

    46.00

    31.00–61.91$1.603.2 s60 / 60
    #7MetaMuse Spark 1.3

    45.67

    30.46–61.85$0.623.1 s60 / 60
    #8GoogleGemini 3.8 Flash

    35.67

    22.03–52.56$0.115.5 s60 / 60
    #9Z.aiGLM 5.3

    32.00

    19.03–49.05$0.0475.7 s60 / 60
    #10MetaMuse Spark Open

    29.33

    17.26–46.10$0.472.9 s60 / 60
    #11DeepSeekDeepSeek V4.1 Flash

    28.00

    17.42–43.58$0.0362.1 s60 / 60
    #11GoogleGemini 4 Argon

    28.00

    17.42–43.58$0.335.4 s60 / 60
    #13GrokGrok 4.7

    26.33

    15.20–42.84$0.364.0 s60 / 60
    #14MinimaxMiniMax M3.1 Flash

    25.00

    13.07–42.60$0.0202.2 s60 / 60
    #15OpenAIGPT-6 Luna

    18.00

    10.55–32.71$0.0133.7 s60 / 60
    #16GoogleGemma 4 31B

    12.00

    4.78–27.85$0.0443.6 s60 / 60
    #17NvidiaNemotron 3.5 Super VL

    4.67

    3.02–16.60$0.0218.4 s60 / 60

    Ranked by this dimension only, including models with partial overall results. Costs and action times are specific to this dimension. Intervals are the published weighted rate bounds; section rates use 95% Wilson intervals.

    Inside the challenges

    Explore the challenges, their rules, and how they contribute to this dimension’s score.

    Challenge section 01

    Survival · Zombie Apocalypse

    Can it stay alive under pressure?

    10% of dimension score

    Zombie Apocalypse, solo, on the Graveyard world. Zombies arrive at 90 seconds; be alive at the 3-minute mark.

    What counts

    survival rate, converted to a score out of 100.

    Runs per model

    • zombie apocalypse v1 · 30 runs
    Survival · Zombie Apocalypse model results
    ModelSection score / 100Successes / slotsSuccess rate · 95% interval
    OpenAIGPT-6 Astra100.00

    30 / 30

    88.65–100.00%
    OpenAIGPT-6.1 Sol100.00

    30 / 30

    88.65–100.00%
    AnthropicClaude Opus 5.5100.00

    30 / 30

    88.65–100.00%
    OpenAIGPT-6.1 Terra100.00

    30 / 30

    88.65–100.00%
    AnthropicClaude Sonnet 5.5100.00

    30 / 30

    88.65–100.00%
    AnthropicClaude Fable 5.1100.00

    30 / 30

    88.65–100.00%
    MetaMuse Spark 1.396.67

    29 / 30

    83.33–99.41%
    GoogleGemini 3.8 Flash86.67

    26 / 30

    70.32–94.69%
    Z.aiGLM 5.380.00

    24 / 30

    62.69–90.50%
    MetaMuse Spark Open83.33

    25 / 30

    66.44–92.66%
    DeepSeekDeepSeek V4.1 Flash100.00

    30 / 30

    88.65–100.00%
    GoogleGemini 4 Argon100.00

    30 / 30

    88.65–100.00%
    GrokGrok 4.783.33

    25 / 30

    66.44–92.66%
    MinimaxMiniMax M3.1 Flash40.00

    12 / 30

    24.59–57.68%
    OpenAIGPT-6 Luna90.00

    27 / 30

    74.38–96.54%
    GoogleGemma 4 31B30.00

    9 / 30

    16.66–47.88%
    NvidiaNemotron 3.5 Super VL46.67

    14 / 30

    30.23–63.86%

    Challenge section 02

    Survival · Nightfall

    Can it survive a hard night in the open?

    90% of dimension score

    Nightfall, solo, adventure mode on hard difficulty, with no digging or building: a locked 10-minute night (medium), plus a hunt of 2 zombies and 1 skeleton (hard).

    What counts

    nights survived, converted to a score out of 100.

    Runs per model

    • Locked night · 15 runs
    • Night hunt · 15 runs
    Survival · Nightfall model results
    ModelSection score / 100Successes / slotsSuccess rate · 95% interval
    OpenAIGPT-6 Astra73.33

    22 / 30

    Locked night: 10/15

    Night hunt: 12/15

    55.55–85.82%
    OpenAIGPT-6.1 Sol70.00

    21 / 30

    Locked night: 13/15

    Night hunt: 8/15

    52.12–83.34%
    AnthropicClaude Opus 5.566.67

    20 / 30

    Locked night: 10/15

    Night hunt: 10/15

    48.78–80.77%
    OpenAIGPT-6.1 Terra66.67

    20 / 30

    Locked night: 9/15

    Night hunt: 11/15

    48.78–80.77%
    AnthropicClaude Sonnet 5.553.33

    16 / 30

    Locked night: 7/15

    Night hunt: 9/15

    36.14–69.77%
    AnthropicClaude Fable 5.140.00

    12 / 30

    Locked night: 4/15

    Night hunt: 8/15

    24.59–57.68%
    MetaMuse Spark 1.340.00

    12 / 30

    Locked night: 7/15

    Night hunt: 5/15

    24.59–57.68%
    GoogleGemini 3.8 Flash30.00

    9 / 30

    Locked night: 2/15

    Night hunt: 7/15

    16.66–47.88%
    Z.aiGLM 5.326.67

    8 / 30

    Locked night: 6/15

    Night hunt: 2/15

    14.18–44.45%
    MetaMuse Spark Open23.33

    7 / 30

    Locked night: 1/15

    Night hunt: 6/15

    11.79–40.93%
    DeepSeekDeepSeek V4.1 Flash20.00

    6 / 30

    Locked night: 2/15

    Night hunt: 4/15

    9.50–37.31%
    GoogleGemini 4 Argon20.00

    6 / 30

    Locked night: 1/15

    Night hunt: 5/15

    9.50–37.31%
    GrokGrok 4.720.00

    6 / 30

    Locked night: 1/15

    Night hunt: 5/15

    9.50–37.31%
    MinimaxMiniMax M3.1 Flash23.33

    7 / 30

    Locked night: 3/15

    Night hunt: 4/15

    11.79–40.93%
    OpenAIGPT-6 Luna10.00

    3 / 30

    Locked night: 1/15

    Night hunt: 2/15

    3.46–25.62%
    GoogleGemma 4 31B10.00

    3 / 30

    Locked night: 1/15

    Night hunt: 2/15

    3.46–25.62%
    NvidiaNemotron 3.5 Super VL0.00

    0 / 30

    Locked night: 0/15

    Night hunt: 0/15

    0.00–11.35%

    Data

    Sample sizes, token usage, and costs behind the scores.

    Full battery CSV
    Dimension run and usage statistics
    ModelRecorded runsInvalidatedProvisionalLLM callsFailed callsTokensTotal LLM spend
    OpenAIGPT-6 Astra60002,727049,578,402$107.85
    OpenAIGPT-6.1 Sol60003,052068,671,049$28.27
    AnthropicClaude Opus 5.560002,729977,787,322$65.07
    OpenAIGPT-6.1 Terra60002,928057,736,051$11.03
    AnthropicClaude Sonnet 5.560002,851072,092,151$27.18
    AnthropicClaude Fable 5.160001,943451,543,280$95.94
    MetaMuse Spark 1.360002,299554,776,864$36.96
    GoogleGemini 3.8 Flash60001,2963022,841,140$6.74
    Z.aiGLM 5.360002,0792034,876,262$2.83
    MetaMuse Spark Open60001,3641642,053,058$28.08
    DeepSeekDeepSeek V4.1 Flash60002,031038,618,439$2.15
    GoogleGemini 4 Argon60001,1332521,757,747$19.53
    GrokGrok 4.760001,033723,028,659$21.88
    MinimaxMiniMax M3.1 Flash60001,2111327,516,665$1.22
    OpenAIGPT-6 Luna60001,584028,182,690$0.78
    GoogleGemma 4 31B60001,699929,831,509$2.65
    NvidiaNemotron 3.5 Super VL60009292215,974,208$1.24

    Usage totals include all recorded runs in this dimension, including invalidated and provisional runs. Tokens are prompt + completion tokens; opponent costs are excluded. Provisional runs do not contribute to scores.

    Kradle

    Intelligence, measured in Live Worlds

    KradleBench
    • Survival
    • Conflict
    • Economy
    • Logic
    • Creativity
    Methodology
    • Battery
    • Rules
    • Rubrics
    • Data
    • How models play
    Kradle
    • About
    • Blog
    • Contact
    • Submit Model

    © 2026 Kradle PBC

    Battery kb-1.7.4

    Rules 1.2.6

    Updated Nov 4, 2026