Cybersecurity Research

OpenAI Launches GPT-6.2 Astra and Sol Across ChatGPT Work, Codex and the API

OpenAI · 18 Nov 2026
Key Takeaway Before switching your teams or applications to new AI models, check what systems and data they can access and review those permissions first.

OpenAI has introduced GPT-6.2 Astra and Sol, two new models available across ChatGPT Work, Codex and the API.

According to the announcement, the models bring better professional outputs, improved coding, stronger computer use and greater efficiency, all at a lower cost. The short summary provided does not detail benchmarks, pricing, availability by region or safety testing, so those points remain unconfirmed.

The mention of computer use is the part most relevant to business security teams. Models that can operate software on a user's behalf are different from chat assistants that only produce text, because they can take actions inside the systems they can reach. Combined with coding capability in Codex and access through the API, the release covers both staff-facing tools and applications that businesses build themselves.

Why it matters

Better computer use at lower cost makes it more likely that staff and developers will give AI agents access to real systems, which widens the set of actions an AI tool can take inside a business. Security leaders should assume adoption will rise quickly across ChatGPT Work, Codex and API integrations, and that access controls set for earlier models may not suit more capable ones.

What to do

  • Review which accounts, files and applications your AI tools and API keys can reach.
  • Check how the new models are enabled in your ChatGPT Work and Codex settings before wide rollout.
  • Test the new models on low-risk tasks before using them with sensitive data or system access.
Running AI agents? Eight controls for securing them, mapped to the ISM ->

Summarised by CISO AI from OpenAI, written with Claude Sonnet 5.5. We link back to every original so you can read it yourself.