OpenAI launches GPT-5.6 model family with tiered pricing
OpenAI made its GPT-5.6 family of models publicly available on July 23, 2026, ending a rollout delayed by U.S. government requests for safety review. The three variants—Sol, Terra, and Luna—are now accessible across ChatGPT, Codex, and the OpenAI API, with pricing differentiated by capability and cost per 1 million tokens OpenAI.
Model Lineup and Pricing
OpenAI positioned the models as a tiered offering for different use cases. Sol, the most capable variant, costs $5 for input tokens and $30 for output tokens per 1M. Terra, the mid-tier option, is priced at $2.50 input and $15 output per 1M. Luna, the most economical, costs $1 input and $6 output per 1M OpenAI. Access varies by subscription tier: Free and Go users default to GPT-5.6 Terra, while Plus, Pro, Business, and Enterprise subscribers can select among all three models.
Phased Rollout and Government Coordination
The public launch follows weeks of negotiation between OpenAI and the White House. On July 7, 2026, Reuters reported OpenAI would release its "most advanced AI model" after a delay prompted by government requests. A June 26 press release from OpenAI indicated the company agreed to a staggered release timeline, stating it was "actively working with the administration on safety testing and reviews before deploying GPT-5.6 widely," according to Politico.
White House officials pushed back on characterizations of formal "approval," with Politico reporting that one official clarified no such permission was required or granted. OpenAI's statement, however, signaled tension over the process: the company said it did not believe "this kind of government access process should become the long-term default."
Gradual Availability
OpenAI said the rollout "is starting globally now and will continue gradually toward full availability over the next 24 hours" OpenAI. This phased approach mirrors enterprise release patterns for high-impact infrastructure and suggests the company is managing load and monitoring system performance during initial deployment.