GPT-6 Astra beat Portal start to finish without human help in under 24 hours

OpenAI's GPT-6 Astra model autonomously completed the entire game Portal in 23 hours and 43 minutes without human assistance after the initial goal was set, according to developer cozyblaze. The run used at least $570 in token usage at Astra's list price, though cozyblaze utilized a $200 Codex subscription. The agent controlled the game via MCP and a modified SourcePauseTool, pausing to analyze screenshots and player data before resuming.

GPT-6 Astra beat Portal start to finish without human help in under 24 hours GPT-6 Astra has played through all of Portal on its own. Developer cozyblaze posted on X https://x.com/cozyblazex/status/2096383114851533097 that the model navigated the full game and reached the credits without any human help after the initial goal was set. The run took about 23 hours and 43 minutes. Token usage https://github.com/cozyblaze/portal-agent/blob/main/evidence/summary.json adds up to at least $570 at Astra's list price, though cozyblaze used a $200 Codex subscription. The model controls Portal through MCP and a modified SourcePauseTool that pauses the game while it thinks. During each pause, the agent sees screenshots, player position, and camera angle, picks its inputs, then lets the game resume. The video below has those pauses cut out. Code and docs are on GitHub https://github.com/cozyblaze/portal-agent/blob/main/docs/how-it-works.md . Cozyblaze noted that OpenAI set a goal in 2016 to solve many different games with a single agent https://the-decoder.com/openais-latest-agent-brings-its-long-standing-2017-vision-for-advanced-ai-capabilities-closer-to-reality/ . Problems remain, but watching an agent beat a full game on its own is a small taste of that original vision https://x.com/cozyblazex/status/2096383117112262864 . GPT-6 Astra is "the worst model we'll ever get," https://x.com/cozyblazex/status/2096383118580351281 cozyblaze added. AI News Without the Hype – Curated by Humans Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section. Subscribe now