OpenAI GPT-5.6 Sol, Terra, and Luna Models Publicly Launched on July 9, Previously Restricted by Government Requirements

OpenAI will publicly launch the GPT-5.6 Sol, Terra, and Luna models on July 9, 2026. Previously, OpenAI had submitted the models to the government for safety evaluation, and only expanded global preview access after receiving approval from the Trump administration.

At the end of June 2026, OpenAI initially only opened these three models to a small group of trusted partners. This was because the AI cybersecurity executive order signed by the Trump administration in early June required major developers to voluntarily submit their strongest models for review 30 days before public release. OpenAI stated at the time that it did not believe such a government access process should become a long-term default practice, but cooperated to achieve public release as quickly as possible. The AI Standards and Innovation Center of the Department of Commerce then completed additional testing, and OpenAI sent technical experts to Washington to answer questions on-site.

Model Positioning and Pricing Details

OpenAI's descriptions of the three models clearly distinguish their use cases. Sol is the company's most capable model for agentic tasks to date, targeting coding, biology, and cybersecurity. Terra is a balanced model for daily work, with performance similar to GPT-5.5 but at half the cost. Luna is the company's lowest-cost version, offering strong capabilities at the lowest price. Specific pricing: Sol input $5 per million tokens, output $30; Terra input $2.5, output $15; Luna input $1, output $6.

These prices directly determine the applicable boundaries for different scenarios. Terra costs only half of GPT-5.5 for the same performance, making it clearly attractive for enterprises that need a large amount of daily reasoning. Luna further lowers the barrier, suitable for high-frequency, low-complexity tasks.

Impact of Government Review on Release Timeline

The government review did not result in a 30-day waiting period. OpenAI announced on June 26 that it had completed initial evaluations in partnership with government entities, received approval to expand access on July 8, and publicly launched on July 9. The entire process from restricted release to approval took only about two weeks, demonstrating efficient coordination in testing and communication between the two sides.

During the same period, Anthropic went through a similar process. The company launched the Fable 5 and Mythos 5 models on June 9, temporarily suspended access for foreign users due to government export controls, and regained approval to restore access on June 30. Both companies were first restricted in access under government requirements, then approved to expand, indicating that current policy review for the strongest models has become a short-term norm.

Practical Impact on Developers and Enterprises

Starting July 9, developers can directly call the three models without needing to go through restricted partner channels. Sol is suitable for coding or security research projects requiring strong agentic capabilities. Terra is suitable for routine application development to control costs. Luna is suitable for high-concurrency, low-cost scenarios. Enterprises can match models by task complexity: choose Sol for high-value biology or cybersecurity tasks, Terra for daily office work, and Luna for large-scale, low-complexity calls.

With transparent pricing, enterprises can more accurately calculate long-term usage costs. Terra's halved pricing means that with the same budget, they can run twice as many inference tasks, which is particularly crucial for budget-sensitive startups and mid-sized companies.