OpenAI Releases Limited Preview of GPT-5.6 Model Series

OpenAI launched a limited preview of its GPT-5.6 model series on June 26, 2026, to a small group of trusted partners. The release includes three variants: Sol, the strongest model; Terra, which matches GPT-5.5 performance at half the cost; and Luna, the lowest-cost option. Sol is priced at $5 per million input tokens and $30 per million output tokens.

Terra costs $2 per million input tokens.

The company gave the U.S. government a preview of GPT-5.6 before the limited release. GPT-5.6 introduces a max reasoning effort mode that gives Sol more time to reason deeply.

Trump signed an AI cybersecurity order earlier in June requiring companies to present their most powerful models for voluntary government review 30 days before public release. OpenAI stated it does not believe government access processes should become the long-term default for model releases.

OpenAI spent several weeks finding weaknesses in the models and dedicated 700,000 GPU hours to finding universal jailbreaks. The company trained the models to refuse prohibited cyber assistance and pledged a rapid-response process for newly discovered jailbreaks.

Anthropic suspended access to its Mythos 5 and Fable 5 models a couple of weeks earlier after a U.S. government directive. Amazon and other companies had notified authorities that the models could be jailbroken and used for malicious purposes. The government later permitted Anthropic to redeploy Mythos to a select group of organizations.

OpenAI revised its preparedness framework in April 2026. The company dedicated about 700,000 A100e GPU hours to automated red-teaming of GPT-5.6.