OpenAI launches the GPT-5.6 model family

OpenAI released the GPT-5.6 family on July 9, 2026, moving it from a limited preview to general availability. Rather than introducing a single model, the company is offering three capability tiers. Sol is the flagship for the most demanding work, Terra is the balanced lower-cost model for everyday work, and Luna is the fastest and most affordable option. OpenAI says the family is intended to deliver more useful work per token, while letting users make a clearer choice between capability, speed and cost.

GPT-5.6 is available in ChatGPT, Codex and the OpenAI API, with OpenAI saying the global rollout would continue toward full availability over 24 hours. In ChatGPT Work and Codex, Free and Go users get Terra. Plus, Pro, Business and Enterprise customers can choose Sol, Terra or Luna and set an effort level for each. Pro and Enterprise users can also select Sol Pro in ChatGPT for complex work. This makes the release relevant to casual users as well as teams that need a more capable model for selected tasks.

The company has also introduced higher settings for difficult workflows. `max` gives GPT-5.6 more time to reason, explore alternatives and check its work. `ultra`, OpenAI's highest-capability setting, coordinates four agents in parallel by default. OpenAI presents that setting as a trade-off: it uses more tokens, but is designed to deliver stronger results and reduce the time needed for demanding tasks.

For makers, all three models are accessible through the API. OpenAI says Programmatic Tool Calling in the Responses API lets GPT-5.6 write and run in-memory programs that coordinate tools and process intermediate results. A multi-agent feature is initially in beta. The published prices per million tokens are $5 input and $30 output for Sol, $2.50 and $15 for Terra, and $1 and $6 for Luna.

OpenAI says it also tested the models and their safeguards extensively before release, combining human red teaming with large-scale automated testing. Its broader market move is clear: model providers are increasingly selling portfolios rather than one universal flagship. Users, developers and businesses will need to decide which tier fits a task, then test that choice against their own quality, cost and safety requirements.