Skate 3 Bench
A skateboarding game generated entirely from scratch by Theo using Claude, iteratively optimized through screenshots and scoring by a panel of LLM reviewers.
Timeline (last 90 days)
Theo built projects on Claude including TypeScript rewritten in Rust (TS Rust), an alternative compiler that accelerates Rust compilation a hundredfold (Lust), and a skateboarding game Skate 3 Bench entirely generated by Claude, cumulatively running Claude for close to 4000 hours.
In Skate 3 Bench, after the model is satisfied with a version it automatically takes a screenshot, then feeds the screenshot together with the corresponding frame from the original Skate to another LLM review panel for scoring, and continues iterating based on the feedback.
TS Rust cost about $38,000 at API prices, while Skate 3 Bench's inference cost approached $5,000, consuming over 12 billion tokens.
All Facts (3)
Theo built projects on Claude including TypeScript rewritten in Rust (TS Rust), an alternative compiler that accelerates Rust compilation a hundredfold (Lust), and a skateboarding game Skate 3 Bench entirely generated by Claude, cumulatively running Claude for close to 4000 hours.
50%UnverifiedIn Skate 3 Bench, after the model is satisfied with a version it automatically takes a screenshot, then feeds the screenshot together with the corresponding frame from the original Skate to another LLM review panel for scoring, and continues iterating based on the feedback.
50%UnverifiedTS Rust cost about $38,000 at API prices, while Skate 3 Bench's inference cost approached $5,000, consuming over 12 billion tokens.
50%