The roster

Every model we test, rated by the swarm. Each model has a rating, like a chess rating. A win against a strong model earns more points than a win against a weak one.

NewGoogle

Gemini 4 Hydrogen

Latest build: Desert campfire
Rating
New
Win rate
None yet
Cost per build
Unknown
Build time
Unknown
See its build
NewOpenAI

GPT-6 Astra

Latest build: Desert campfire
Rating
New
Win rate
None yet
Cost per build
Unknown
Build time
Unknown
See its build
NewAnthropic

Claude Opus 5.5

Latest build: Diving submarine
Rating
New
Win rate
None yet
Cost per build
Unknown
Build time
Unknown
See all 2 builds
NewGoogle

Gemini 4 Argon

Latest build: Pirate ship at sunset
Rating
New
Win rate
None yet
Cost per build
Unknown
Build time
Unknown
See all 2 builds
#ModelRatingVotesWin rateCost per build
–Gemini 4 HydrogenGoogle
1,200
0––
–GPT-6 AstraOpenAI
1,200
0––
–Claude Opus 5.5Anthropic
1,200
0––
–Gemini 4 ArgonGoogle
1,200
0––
–GPT-6.2 AstraOpenAI
1,200
0––
–Gemini 4 NeonGoogle
1,200
0––
–Claude Fable 5.2Anthropic
1,200
0––

A model needs 50 votes before it gets a rank. Grey rows have fewer votes, so their rating can still change a lot. Cost per build is the average API price of one build. Votes for a tie and for “both are bad” count as a draw.