OpenAI unveils GPT-5.6 with limited preview for trusted partners

OpenAI has launched a limited preview of its new GPT-5.6 model family. The lineup includes three variants: Sol, Terra, and Luna.

The flagship model, GPT-5.6 Sol, is described by OpenAI as its most capable model yet. According to the company, it delivers stronger performance in coding, biology, and cybersecurity.

GPT-5.6 also introduces two new reasoning modes:

• max for deeper analysis;
• ultra, which uses subagents to tackle more complex tasks.

GPT-5.6 Terra is positioned as a balanced model for everyday use, while Luna is the faster and more affordable option. OpenAI says Terra offers performance comparable to GPT-5.5 at half the cost, while Luna delivers "strong capability" at the lowest price point in the lineup.

"For coding workflows, GPT-5.6 Sol sets a new state of the art on Terminal-Bench 2.1, which tests command-line workflows requiring planning, iteration, and tool coordination," the company said.

The rollout, however, is not yet public. At launch, GPT-5.6 will only be available through the API and Codex to a limited group of trusted partners and organizations.

• OpenAI says the decision follows a request from the U.S. government. The company shared its plans and model capabilities with officials in advance and agreed to begin with a limited preview.
• OpenAI also says it does not want this approach to become standard practice.
• According to the company, restricted access delays developers, businesses, cybersecurity professionals, and other users from getting access to its most advanced tools.

OpenAI says it agreed to the temporary rollout in order to help shape future rules governing the release of increasingly powerful AI models.

The company also placed a strong emphasis on safety. OpenAI says GPT-5.6 is its most capable cybersecurity model to date, while still remaining below the company's internal Cyber Critical risk threshold. In testing, the model was able to identify vulnerabilities and individual exploit components but was not capable of autonomously building a complete attack chain.

OpenAI has also published API pricing. GPT-5.6 Sol costs $5 per million input tokens and $30 per million output tokens, while Terra is priced at $2.50/$15 and Luna at $1/$6. The company has also updated its prompt caching policy: cached prompt writes for GPT-5.6 and future models will cost 1.25× the standard input rate, while cached reads will continue to receive a 90% discount.

Broader availability across ChatGPT, Codex, and the API is expected to roll out over the coming weeks.