01 What happened

OpenAI released GPT-6 Astra, a model demonstrating high performance across key benchmarks including FrontierMath and ARC-AGI-3, while introducing new alignment safeguards.

02 Key details

  • OpenAI reports scores of 97.6% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on ExploitBench.
  • In an OSWorld 2.0 simulation, Astra completed tasks in about 40 minutes versus 75 minutes for GPT-5.6 Sol, with a higher task score.
  • In one test of crossing an authorized task boundary, Astra did so in 0% of cases versus 48% for GPT-5.6 Sol without production safeguards.
  • OpenAI reports that the model can identify previously unknown vulnerabilities, increasing the need for controls on its use.

03 Why it matters

This release marks a shift toward highly autonomous technical workflows, highlighting the dual-use nature of models capable of both vulnerability discovery and defense.

04 Who it matters to

Machine learning engineers, software developers, information security specialists and enterprise AI customers.

Original source
OpenAI