Hacker News
Nineteen Model Configurations Played 171 Rounds of StarCraft: Brood War and None Beat a Basic Human
Ben Swerdlow built a version of Brood War playable only through agents and ran a full round-robin of 19 model/effort configurations across 171 matches, publishing results on September 19. Codex Astra at xhigh effort went 18-0 for a 100% win rate, Astra at medium 16-2, Claude Fable 15-3 at 83.3% and Astra at low 14-4, while Grok 4.6 at xhigh managed only six command batches across 43 minutes. Styles diverged legibly, with Claude playing textbook macro and Codex favoring unconventional trick strategies, and the ceiling is the finding: no configuration beat a basic human player.
↳ Follow the thread