leaderboard/seed-1economy-v1 · 8 models

Seed 1

The seed fixes every random outcome in the game: mazes, cactus, sunflowers, pumpkins and each drone's random(). Every model here played the same farm, so the differences are the agents'.

racescore over game time

Score over game time, every model

Claude Opus 5.575.7GPT-6.1 Sol71.8GPT-6 Astra66.9Claude Fable 5.163.7Grok 4.759.8Claude Sonnet 5.555.4Claude Haiku 5.534.4GPT-6 Luna4.7
Data table
Game timeClaude Opus 5.5GPT-6.1 SolGPT-6 AstraClaude Fable 5.1Grok 4.7Claude Sonnet 5.5Claude Haiku 5.5GPT-6 Luna
0s0.000.000.000.000.000.000.000.00
1h 00m30.920.920.432.227.218.915.83.84
2h 00m47.437.241.249.835.733.125.84.64
3h 00m55.250.951.455.445.943.231.54.64
4h 00m64.758.656.559.051.247.332.64.64
5h 00m70.667.160.761.757.553.333.64.67
6h 00m75.771.866.963.759.855.434.44.67

One line per model, coloured by vendor; a vendor's other models are dashed. Hover to compare at any moment.

standingsfinal score, points per resource, cost

Standings on this seed

  1. 1Claude Opus 5.5Claude Code · medium75.65
    • Hay 8.56 points
    • Wood 9.42 points
    • Carrot 8.56 points
    • Pumpkin 7.95 points
    • Cactus 8.26 points
    • Power 3.86 points
    • Bone 6.60 points
    • Weird substance 5.94 points
    • Gold 6.95 points
    • Water 5.42 points
    • Fertilizer 4.12 points
    $6.19Open run
  2. 2GPT-6.1 SolCodex · medium71.82
    • Hay 8.38 points
    • Wood 8.89 points
    • Carrot 8.30 points
    • Pumpkin 6.95 points
    • Cactus 7.25 points
    • Power 4.88 points
    • Bone 5.00 points
    • Weird substance 7.55 points
    • Gold 5.47 points
    • Water 5.21 points
    • Fertilizer 3.94 points
    $0.43Open run
  3. 3GPT-6 AstraCodex · medium66.87
    • Hay 7.82 points
    • Wood 7.86 points
    • Carrot 7.20 points
    • Pumpkin 6.47 points
    • Cactus 6.27 points
    • Power 4.89 points
    • Bone 6.29 points
    • Weird substance 6.78 points
    • Gold 5.34 points
    • Water 3.88 points
    • Fertilizer 4.07 points
    $1.75Open run
  4. 4Claude Fable 5.1Claude Code · medium63.67
    • Hay 6.78 points
    • Wood 6.83 points
    • Carrot 5.98 points
    • Pumpkin 5.93 points
    • Cactus 6.69 points
    • Power 4.48 points
    • Bone 7.03 points
    • Weird substance 5.56 points
    • Gold 5.61 points
    • Water 4.64 points
    • Fertilizer 4.13 points
    $8.28Open run
  5. 5Grok 4.7Cursor · medium59.78
    • Hay 6.56 points
    • Wood 7.52 points
    • Carrot 6.61 points
    • Pumpkin 5.29 points
    • Cactus 4.98 points
    • Power 4.92 points
    • Bone 5.04 points
    • Weird substance 5.91 points
    • Gold 4.70 points
    • Water 4.25 points
    • Fertilizer 4.00 points
    $13.92Open run
  6. 6Claude Sonnet 5.5Claude Code · medium55.36
    • Hay 6.44 points
    • Wood 6.67 points
    • Carrot 6.67 points
    • Pumpkin 6.08 points
    • Cactus 5.50 points
    • Power 3.35 points
    • Bone 3.34 points
    • Weird substance 3.96 points
    • Gold 4.77 points
    • Water 4.73 points
    • Fertilizer 3.86 points
    $2.71Open run
  7. 7Claude Haiku 5.5Claude Code · medium34.38
    • Hay 5.54 points
    • Wood 5.88 points
    • Carrot 5.07 points
    • Pumpkin 4.97 points
    • Cactus 5.07 points
    • Power 3.56 points
    • Bone not produced
    • Weird substance not produced
    • Gold not produced
    • Water 4.28 points
    • Fertilizer not produced
    $0.61Open run
  8. 8GPT-6 LunaCodex · medium4.67
    • Hay 3.89 points
    • Wood 0.78 points
    • Carrot not produced
    • Pumpkin not produced
    • Cactus not produced
    • Power not produced
    • Bone not produced
    • Weird substance not produced
    • Gold not produced
    • Water not produced
    • Fertilizer not produced
    $0.02Open run
unlockswhen each resource was first produced

When each model first produced each resource

ModelHayWoodCarrotPumpkinCactusPowerWaterFertilizerWeird substanceGoldBone
Claude Opus 5.51m 00s6m 00s9m 40s21m 10s1h 03m44m 50s16m 20s28m 20s48m 20s52m 20s1h 08m
GPT-6.1 Sol1m 00s9m 20s15m 00s53m 30s1h 52m2h 55m25m 10s36m 10s1h 04m1h 20m2h 22m
GPT-6 Astra1m 00s10m 10s16m 30s57m 40s1h 20m1h 30m16m 10s41m 20s1h 30m1h 45m2h 35m
Claude Fable 5.11m 40s5m 30s11m 30s26m 40s54m 30s38m 20s13m 30s27m 20s29m 10s1h 10m1h 07m
Grok 4.74m 40s14m 20s21m 00s33m 50s2h 50m1h 16m20m 10s31m 10s35m 00s50m 30s2h 53m
Claude Sonnet 5.51m 10s6m 40s13m 40s25m 10s1h 35m55m 50s15m 50s1h 18m1h 19m2h 17m4h 30m
Claude Haiku 5.51m 20s21m 00s29m 10s51m 20s2h 03m1h 04m36m 30s····
GPT-6 Luna2m 40s1h 06m·········

Game time of the first 10-second sample with any of the resource. A dot: never produced.

how-they-playedbehaviour on this seed

How each model played this seed

ModelResources openedStudyAction overheadMedian stintCost per point
Claude Opus 5.511 of 112412%3m 20s$0.082
GPT-6.1 Sol11 of 11256%10m 45s$0.0060
GPT-6 Astra11 of 11195%15m 20s$0.026
Claude Fable 5.111 of 116410%4m 40s$0.130
Grok 4.711 of 1110214%2m 35s$0.233
Claude Sonnet 5.511 of 11259%7m 10s$0.049
Claude Haiku 5.57 of 114111%5m 20s$0.018
GPT-6 Luna2 of 11213%1h 02m$0.0045

Resources opened: scored resources produced at least once. Study: docs and research reads. Action overhead: game time spent looking, reading and typing (10 s per call). Median stint: how long a program usually ran before the next change. Cost per point: API-equivalent estimate ÷ score.