Key facts
- OpenAI is delaying the broad release of its GPT 5.6 AI model series.
- The U.S. government requested the staggered release due to concerns about the model's capabilities.
- GPT 5.6 includes three versions: Sol, Terra, and Luna, with Sol being the most advanced.
- OpenAI expressed concerns that government intervention in AI model releases should not become standard practice.
- The company has also been updating and retiring older models like GPT-5.5 and GPT-4.5.
OpenAI is staggering the release of its latest AI model, GPT 5.6, following a request from the U.S. government. This move, which OpenAI has expressed dissatisfaction with, mirrors a similar situation with Anthropic's Mythos product. The company stated that such delays prevent essential AI tools from reaching users, developers, and cyber defenders.
OpenAI announced on Friday that it is launching the GPT 5.6 series with a limited preview for a small group of trusted, U.S.-based partners. This decision was made after discussions with government agencies, including the Office of the National Cyber Director and the Office of Science and Technology Policy. OpenAI's CEO, Sam Altman, informed staff that the government would approve access on a customer-by-customer basis during this preview period, with a broader release expected in the coming weeks if the process is successful.
The GPT 5.6 suite comprises three versions: Sol, the most powerful; Terra, with slightly lower performance and cost; and Luna, the lowest-cost option. OpenAI described Sol as its strongest model yet, emphasizing that it is better at identifying and fixing vulnerabilities than at carrying out end-to-end attacks, and does not cross an internal "cyber critical threshold."
In parallel developments, OpenAI has been updating its existing models. GPT-5.5 Instant has been improved for response style and quality, with writing and coding functionality now directly supported in chat responses. Older models like OpenAI o3 and GPT-4.5 are being retired from ChatGPT, with sunset periods in place. GPT-5.4 mini has been rolled out as a fallback for paid users experiencing rate limits with GPT-5.4 Thinking, and GPT-5.4 Thinking itself has been enhanced with improved reasoning, coding, and agentic workflows.