StarSkirmish Bench v0.2: LLMs Write StarCraft Bots in One Hour

1 reports archived1 independent sourcesupdated 10/16/2026, 22:18:44

Synthesis & Latest Updates

1 Sources Cross-Validated

StarSkirmish Bench v0.2 evaluates flagship LLMs on their ability to write competitive StarCraft: Brood War bots in C++ within a one-hour limit. The benchmark tests long-horizon reasoning and agentic coding, with GPT-6 Astra leading at a score of 57 against human-written bot baselines.

LATEST/StarSkirmish Bench v0.2 evaluates flagship LLMs on their ability to write competitive StarCraft: Brood War bots in C++ within a one-hour limit. The benchmark tests long-horizon reasoning and agentic coding, with GPT-6 Astra leading at a score of 57 against human-written bot baselines.

HEAT TRENDHourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 1 reports · Latest first

Hacker News AIT2·Oct 16, 22:1868 pts
StarSkirmish Bench v0.2: LLMs Write StarCraft Bots in One Hour
- StarSkirmish v0.2 requires LLMs to write Protoss bots in C++ within 1 hour and compete against human-coded bots.
- Scores reflect average win rates (0-100), with GPT-6 Astra scoring 57, highlighting current top-tier agentic coding performance.
- The benchmark specifically targets long-horizon reasoning and the ability to learn from practice games to improve strategy.
View Related Reports & Analysis Original Source