OpenAI is releasing GPT-5.6 in three models, from the most powerful to the most economical.

OpenAI releases GPT-5.6 in Sol, Terra, and Luna models to optimize token efficiency and run multi-agent workflows via ChatGPT, Codex, and its API today.

OpenAI is releasing GPT-5.6 generation from preview, offering it in three models with distinct roles: Sol, the flagship; Terra, a balanced model for everyday work; and Luna, the fastest and most economical. The number denotes the generation, while Sol, Terra, and Luna become durable capability tiers, each expected to evolve at its own pace. The central argument is efficiency: getting more out of each token for a better result-to-cost ratio.

Several new features target agentic uses. A setting called "ultra" coordinates four agents in parallel by default, up to sixteen, to accelerate heavy tasks, while a "max" setting allows for more reasoning time. On the API side, Programmatic Tool Calling lets the model write and execute small programs that orchestrate tools and filter intermediate data, reducing round trips, and a multi-agent mode is arriving in beta.

OpenAI also highlights a leap in design judgment: thanks to better computer usage, the model inspects and corrects an interface's rendering rather than just producing its code. For office work, it transforms scattered context (Slack, Notion, Microsoft 365, Google Drive) into editable presentations, documents, and spreadsheets, capable of reproducing the graphic system of an existing slide deck.

The publisher also claims progress in cyber and science, with its defensive cybersecurity capabilities (code review, patches, threat modeling) reserved for verified users via its Daybreak Trusted Access for Cyber program, with restrictions for at-risk entities and jurisdictions. On safety, OpenAI says it has conducted its most extensive evaluations to date, approximately 700,000 GPU hours of automated red teaming, and estimates that the models do not cross the "critical" threshold in either biology or cybersecurity.

GPT-5.6 is available today in ChatGPT, Codex, and the API, with global deployment rolling out over twenty-four hours. Pricing per million tokens varies by model: five dollars for input and thirty for output for Sol, two dollars fifty and fifteen for Terra, one dollar and six for Luna.