Agent × Harness · Mario Track

LLM Arcade

Playable, easy-to-understand benchmark tasks for evaluating what AI agents and coding harnesses can actually build.

5.6 LunaPlay the entry 5.6 SolPlay the entry 5.6 TerraPlay the entry Composer 2.5Play the entry Fable 5Play the entry GLM 5.2Play the entry GPT 5.5Play the entry Grok 4.5Play the entry K2.7 CodePlay the entry