Field study / Version 1 · Archive
Version 1 · Archive

ChronoBench

Measuring how far different language models can progress inChrono Trigger using an autonomous vision-based game agent.

Last updated: 2026-04-18 · Frozen at 17 runs.

These runs were produced by the v1 harness. For current runs on the v3 architecture, seeVersion 3, or browse the intermediatev2 archive.

Leaderboard

ModelProviderLast Checkpoint123456CyclesStuckTokens Inms/cycleEst. CostDate
google/gemma-4-26b-a4bLocalmarle_met182448180——200*492,018,37224,398Local · $02026-04-11
google/gemini-3-flash-previewOpenRoutermarle_met24307290——200*523,726,27713,687$2.052026-04-16
google/gemini-3-flash-previewOpenRoutermarle_met142172137——200*534,465,03112,932$2.422026-04-17
google/gemini-3-flash-previewOpenRoutermarle_met15183369——200*684,446,61914,076$2.412026-04-17
google/gemini-3-flash-previewOpenRoutermarle_met14425174——200*334,576,34414,267$2.472026-04-17
google/gemini-3-flash-previewLocalmarle_met14639299——200*683,158,12413,429Local · $02026-04-18
google/gemma-4-26b-a4bLocalfair_entered184866———100*311,055,47025,174Local · $02026-04-09
google/gemma-4-26b-a4bLocalfair_entered182436———100*421,062,86225,482Local · $02026-04-09
google/gemma-4-26b-a4bLocalfair_entered244860———200*502,077,87624,953Local · $02026-04-11
google/gemma-4-26b-a4bLocalfair_entered243084———200*492,391,20923,392Local · $02026-04-12

* = cycle budget exhausted

Cycles per checkpoint

04080119159199CyclesLeft Bedroomgoogle/gemma-4-26b-a4b: 1818google/gemini-3-flash-preview: 2424google/gemma-4-31b: 2424Exited Housegoogle/gemma-4-26b-a4b: 2424google/gemini-3-flash-preview: 3030google/gemma-4-31b: 4848Reached the Fairgoogle/gemma-4-26b-a4b: 4848google/gemini-3-flash-preview: 7272google/gemma-4-31b: 6666Met Marlegoogle/gemma-4-26b-a4b: 180180google/gemini-3-flash-preview: 9090Reached TelepodTime Traveledgoogle/gemma-4-26b-a4bgoogle/gemini-3-flash-previewgoogle/gemma-4-31b

Estimated cost per checkpoint

$0.00$0.40$0.80$1.20$1.60$2.00Cost ($)Left Bedroomgoogle/gemma-4-26b-a4b: $0.00google/gemini-3-flash-preview: $0.25$0.25google/gemma-4-31b: $0.00Exited Housegoogle/gemma-4-26b-a4b: $0.00google/gemini-3-flash-preview: $0.31$0.31google/gemma-4-31b: $0.00Reached the Fairgoogle/gemma-4-26b-a4b: $0.00google/gemini-3-flash-preview: $0.74$0.74google/gemma-4-31b: $0.00Met Marlegoogle/gemma-4-26b-a4b: $0.00google/gemini-3-flash-preview: $0.92$0.92Reached TelepodTime Traveledgoogle/gemma-4-26b-a4bgoogle/gemini-3-flash-previewgoogle/gemma-4-31b

All runs

ModelLast checkpointCyclesStuckDate
google/gemma-4-26b-a4bhouse_exit100*432026-04-09
google/gemma-4-26b-a4bfair_entered100*312026-04-09
google/gemma-4-26b-a4bfair_entered100*422026-04-09
claude-haiku-4-5-20251001—100*122026-04-09
google/gemma-4-26b-a4bfair_entered200*502026-04-11
google/gemma-4-26b-a4bmarle_met200*492026-04-11
google/gemma-4-26b-a4bfair_entered200*492026-04-12
google/gemma-4-26b-a4bfair_entered200*542026-04-12
google/gemma-4-31bfair_entered200*602026-04-12
google/gemini-3-flash-preview—102026-04-16
google/gemini-3-flash-previewmarle_met200*522026-04-16
google/gemini-3-flash-previewmarle_met200*532026-04-17
google/gemini-3-flash-previewmarle_met200*682026-04-17
google/gemini-3-flash-previewmarle_met200*332026-04-17
google/gemini-3-flash-preview—91372026-04-17
google/gemini-3-flash-previewmarle_met200*682026-04-18
google/gemma-4-26b-a4bfair_entered200*312026-04-18