SignalTech/AI
Show HN: Gemini 4 Argon plays StarCraft
Gemini 4 Argon plays StarCraft, signaling new AI gaming capability.
Context
What happened: StarSkirmish Bench v0.2 is a benchmark where LLMs get one hour to write a Protoss bot in C++ for StarCraft: Brood War, then their bots are scored by win rate against human written bots.
Key numbers: GPT-6 Astra tops v0.2; GPT-6.1 Sol is second at just over a tenth of Astra's cost per run. Gemini 4 Argon, Claude Opus 5.5 and Claude Fable 5.1 score between 37 and 39.
Why it matters: The strongest human written bots remain well ahead of every model, and the top predictor of a model's score is how much it practiced against the 7 strongest roster bots.
What changed
StarSkirmish Bench v0.2 is a benchmark where LLMs get one hour to write a Protoss bot in C++ for StarCraft: Brood War, then their bots are scored by win rate against human written bots.
Why it matters
The strongest human written bots remain well ahead of every model, and the top predictor of a model's score is how much it practiced against the 7 strongest roster bots.