Show HN: Gemini 4 Argon plays StarCraft

StarSkirmish Bench v0.2 is a community benchmark where LLMs get one hour in their lab's first-party CLI harness to write a Protoss StarCraft: Brood War bot in C++, scored by win rate against a fixed roster of human-written and demo bots. GPT-6 Astra leads, with GPT-6.1 Sol second at just over a tenth of its cost per run, while the best model still wins only 7% of games against top human bots like Stardust, PurpleWave and BananaBrain. Scores no longer track widely reported evals like the Artificial Analysis Intelligence Index (ρ 0.26) but still correlate with long-horizon coding tests such as…

Read the full story on starskirmish.com

Source

starskirmish.com

All items on the wire