StarSkirmish Bench v0.2 - LLMs write StarCraft bots in one hour

Show HN: Gemini 4 Argon plays StarCraft

StarSkirmish Bench v0.2 challenges LLMs to write a Protoss bot in C++ for StarCraft: Brood War in just one hour. Each model runs in its lab's first-party CLI harness at maximum reasoning effort, then its bot battles human-written bots. The benchmark tests long-horizon reasoning and agentic coding, revealing that top models still struggle against elite human bots. GPT-6 Astra leads with a 57% win rate, while Gemini 4 Argon places third at 39%. The results show no signs of saturation and highlight the value of practice against strong opponents.