← The Polylog AI Intelligence Brief
Morning Edition · Wednesday, September 23, 2026Published at 7:26 AM EDT · New York
A StarCraft Benchmark Finds Agents Can Win Matches Without Learning Strategy
Across 171 round-robin games, a Codex harness running GPT-6 Astra went 18-0 at the highest reasoning setting, yet the benchmark's author says no tested system beat a basic human player.
Ben Swerdlow, co-founder of Freestyle, a company that builds virtual machines for AI agents, converted Blizzard's 1998 real-time strategy game StarCraft: Brood War into an agent benchmark and published the results on September 19. The setup…
Continue the AI Intelligence Brief
Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.
- 5 AI intelligence signals a day
- Frontier labs, compute, and chips
- Model releases and AI infrastructure
- Source-grounded analysis with confidence labels
The Global Intelligence Brief stays free.
Townsquare
Start a discussion in Townsquare.
More from this edition
- OpenAI Halves API Prices With GPT-6 Sol and Luna as Anthropic Cuts Opus Running Costs 40%
- Epoch AI Puts the Fall in the Cost of a Benchmark Answer at 725-Fold in Under 18 Months
- China's Internet Regulator Opens Investigation Into DeepSeek and Moonshot Over Data Routed to Claude
- Trump Tells the United Nations the US Will Call Artificial Intelligence "Super Intelligence"
- OpenAI Forms a Mathematics Advisory Group at the Institute for Advanced Study After Claiming 100 Solved Open Problems
- NVIDIA Ships Isaac ROS 5.0 With Agent Workflows for the Robot Operating System Ecosystem
- OpenAI Publishes Rules for Outside Safety Assessors as Protesters Camp Outside Its Offices
Comments
0No comments yet.