RELEASE·Nov 18, 2026, 1:21 PM ET·Emerging signal

OpenAI releases GPT-6.2 Astra as early testers flag usability problems

TL;DR:

  • OpenAI released GPT-6.2 Astra and GPT-6.2 Sol for professional and other tasks.
  • OpenAI says GPT-6.2 Astra is its first public model to reach its high internal threshold for AI self-improvement.
  • Ethan Mollick praised GPT-6.2 Astra’s simulation output but said it needed more prompting and direction than other models.
  • Dan Shipper criticized GPT-6.2’s writing, coding and design performance relative to Anthropic’s Fable 5.2.

Summary

OpenAI introduced GPT-6.2 Astra and GPT-6.2 Sol, pitching the release for demanding professional work. The company says Astra reaches an internal AI self-improvement threshold that it characterizes as strong assistance for human AI researchers. Early reactions were mixed: Ethan Mollick praised an ambitious simulation result but found the model frustrating to direct, while Dan Shipper criticized several practical skills. Those reactions are user assessments, not a controlled comparison of the models.

Open gizmodo.com in a new tab ↗