DigiNews

Summary

Show HN discusses StarSkirmish Bench v0.2, a benchmark where LLMs write Protoss bots for StarCraft: Brood War within one hour. It compares performance across models (e.g., GPT-6 Astra, Gemini 4 Argon) and analyzes factors like practice against strong bots, cost per run, and run-to-run variance, highlighting how long-horizon coding tasks are evaluated and how models stack up against top human-written bots.