OpenAI unveils GPT-5.6 model: new features, US government involvement and availability dates

On July 9, 2026, OpenAI officially introduced GPT-5.6, calling it its most powerful and functional system. The new generation of the neural network has received a redesigned naming system: the suffixes “Sol”, “Terra” and “Luna” now define three levels of performance.

What’s new in the GPT-5.6 architecture?

The system is now classified into three categories:

• Sol: Flagship model with maximum performance.
• Terra: A balanced option, optimized for daily tasks.
• Luna: The fastest and most affordable level.

Features of implementation and the role of the US government

At the current stage, OpenAI has limited access to the model: GPT-5.6 works exclusively through the Codex API and development environment, and only for a narrow circle of partner organizations approved by the US government. The reason for this step lies in the ability of the neural network to identify vulnerabilities and eliminate bugs in the code, which increases security risks if uncontrolled. Public access to the model as part of ChatGPT has not yet been implemented.

Pricing and access levels

Performance and specialized hardware

The key innovation was hardware optimization. OpenAI announced a partnership with chipmaker Cerebras: GPT-5.6 Sol on their architecture reaches a speed of 750 tokens per second, which significantly exceeds the performance on standard graphics accelerators. Among the new functions, the “max” (for ultra-long logical inferences) and “ultra” (automatic distribution of tasks between agents) modes are also announced.

When will GPT-5.6 be available to users?

A specific release date for ChatGPT has not been announced—the company is targeting the coming weeks. Currently, GPT-5.5 remains the current version for ordinary users.