Benchmarks
How we measure game-development agents.
Two public benchmarks where SEELE02 ranks first, and Oasis — the benchmark we are building for what it takes to ship a game that is actually played.
GameDevBench
333 real Godot game-development tasks, scored by validation. SEELE02-pro leads GPT-5.6 Sol, Claude Opus 4.8 and SEELE02-flash.
65.8%Pass@1 · SEELE02-pro · #1
View results →EvaluatedGameCraft-Bench
140 real Godot game-development tasks, scored by validation. SEELE02-pro ranks first overall.
65.3%Overall · SEELE02-pro · #1
View results →Our benchmarkOasis
Games must be played. Oasis is SeeleAI’s next benchmark for game-development agents — capability-tree-first, Playground-based and runtime-observed.
Read the introduction →