OpenAI on Thursday made its new GPT-5.6 model series publicly available, ending a weeks-long delay imposed by the US government for security vetting. The three-tier family—flagship Sol, mid-range Terra, and low-cost Luna—is now accessible globally, with Terra priced at half the cost of its predecessor. Simultaneously, the company folded its Codex coding tool into the ChatGPT desktop app and launched ChatGPT Work, an AI agent that can operate autonomously for hours to produce documents, spreadsheets, presentations and web applications.

The release followed a review process that OpenAI chief executive Sam Altman described as a “collaborative back and forth” with Trump administration officials. The company made “many changes” after government testers identified potential problems, Altman told CNBC, praising the technical capability of the red-teaming effort. The review was conducted under an executive order that encourages developers of powerful AI models to submit them for pre-release screening. Washington’s intervention reflects a broader push to balance rapid deployment with national security concerns, particularly around the ability of frontier models to identify software vulnerabilities.

The government’s hand has also been felt by Anthropic, OpenAI’s main rival. In June, the Commerce Department briefly imposed export controls on Anthropic’s latest Claude models, Mythos 5 and Fable 5, after a jailbreak was discovered that could expose cyber capabilities. The restrictions were lifted at the end of the month. Both companies are now racing toward initial public offerings. Anthropic has filed confidential paperwork and is targeting a debut this autumn, while OpenAI is considering a listing next year. Secondary-market trading values Anthropic at $1.2 trillion and OpenAI at $908 billion, though shares in the former are so scarce that brokers describe a frenzy of demand with few sellers.

The enterprise push is sharpening competition. Anthropic has its own work agent, Claude Cowork, and both firms are under pressure from corporate customers demanding better value as AI subscription costs rise. OpenAI also introduced GPT-Live, a voice model that can listen and speak simultaneously, delegating complex queries to more powerful systems in the background. Meanwhile, legal headwinds persist: a group of newspapers led by the New York Times asked a federal court on Thursday to sanction OpenAI, alleging it misled the court about its ability to search for copyrighted material in its training data. The next milestones will be the formal IPO filings and the development of a repeatable government approval framework for future model releases.