FrontierThe story, in brief

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

GPT-6 Astra just solved Portal end-to-end without human intervention. That's not a game flex — it's a reasoning milestone.

Illustration of a transparent lens revealing connected networks across layers of paper.
Exploring the next frontier of AI research.AI illustration by KeyNews
The KeyNews take

Why it matters

An autonomous AI system completing a complex puzzle game with zero mid-run human help demonstrates a significant leap in sustained reasoning and planning capability. This is capability evidence, not a product feature, and signals frontier-lab progress on agentic reasoning.

The key facts

5 to know
  1. GPT-6 Astra completed Portal entirely autonomously in ~24 hours

  2. Zero human intervention after initial goal setup

  3. Code and documentation published on GitHub by cozyblaze

  4. Developer characterization: 'the worst model we'll ever get' — suggesting rapid capability ceiling-raising

  5. Portal completion requires sustained multi-step planning, puzzle-solving, and error recovery — not a single-turn task

Go to the source

The Decoderthe-decoder.com

Publisher excerpt: GPT-6 Astra beat the puzzle game Portal entirely on its own in about 24 hours, with zero human help after the initial goal was set. Developer cozyblaze published the code and docs on GitHub. His takeaway: Astra is "the worst model we'll ever get."
Read original report
Back to today's editionMore frontier news

Keep reading

Related stories

More from Frontier