OpenAI releases GPT-5.6 after government approval and unveils ChatGPT Work

On February 10, 2026, OpenAI received approval from the administration of Donald Trump to launch the GPT-5.6 model publicly. For about two weeks beforehand, only organizations cleared by the authorities had been testing it. The same day, the company introduced ChatGPT Work, an AI agent for everyday non-technical tasks built on Codex and the GPT-5.6 family.

Why the launch was delayed

The public release of GPT-5.6 was tied to regulatory restrictions. During the limited preview, the model was offered only to organizations that had received permission from the US government. That testing format lasted roughly two weeks, until OpenAI obtained approval for a wider launch.

The Trump administration's decision removed the key obstacle to the model's public distribution. The original report does not specify which requirements were imposed on GPT-5.6 or which organizations took part in the limited testing.

• The public launch received the Trump administration's approval
• The limited preview lasted about two weeks
• Before approval, the model was available only to government-approved organizations
• OpenAI CEO Sam Altman called GPT-5.6 the company's best model

"This is the best model we have ever created,"

Sam Altman said.

What ChatGPT Work can do

ChatGPT Work combines the capabilities of ChatGPT and Codex to make coding tools useful for people without technical training. The agent is not meant only for programming: OpenAI positions it as a tool for non-technical tasks.

The description provided does not list the full set of ChatGPT Work scenarios and features. All that is known is that the agent uses Codex's capabilities beyond direct software development and runs on the GPT-5.6 family of models.

The new system includes three models: Sol, Terra and Luna. The source material does not say how they differ from one another, what tasks each model handles, or whether users can access them separately.

What this means

OpenAI is widening access to its new flagship model while turning Codex from a tool aimed mainly at developers into the basis of a more general-purpose work agent. The regulatory approval also shows that companies increasingly have to take government bodies' positions into account when launching frontier models publicly.

For users, the main takeaway is that GPT-5.6 is leaving limited preview, while ChatGPT Work offers a way to use Codex's capabilities on tasks that do not require writing code. However, OpenAI has so far disclosed few technical details about the model, the agent, and how roles are split among Sol, Terra and Luna.