After a month under government lock and key, OpenAI's most powerful model is about to go on general sale, and it is walking straight into a fight with Elon Musk. The company said GPT-5.6, its Sol, Terra, and Luna lineup, will roll out broadly on Thursday, roughly two weeks after it agreed to a request from the Trump administration to release the models only to a "small group of trusted partners." The U.S. Department of Commerce, according to Axios, has now cleared the wider launch after further testing and meetings with OpenAI executives. "Happy building," chief executive Sam Altman posted, marking the end of one of the stranger episodes in the AI industry's short history.
The pause was never really about performance. OpenAI describes Sol as its strongest model yet, with gains in coding, biology, and cybersecurity, and it is the cybersecurity part that made Washington nervous. Officials worry that a model good at finding software vulnerabilities could be turned to finding them for a foreign military or intelligence service. That same fear had, weeks earlier, pushed the government to make Anthropic suspend its Fable and Mythos models under an export directive that was lifted only at the end of June. OpenAI complied but made its irritation plain, warning that this kind of government access process should not "become the long-term default" because it "keeps the best tools from users, developers, enterprises, cyber defenders."
No sooner had OpenAI announced its release than Musk moved to steal the moment. A few hours later he said SpaceXAI, the freshly rebranded company formerly known as xAI, would ship its new Grok 4.5 model within days. Musk called it an "Opus-class model, but faster, more token-efficient and lower cost," a direct jab at Anthropic's top tier. The model is reportedly the first built jointly with Cursor, the AI coding startup SpaceX is in the process of buying for a reported 60 billion dollars, and internal tests are said to compare it to Claude Opus 4.8 and OpenAI's GPT-5.5.
Read the two announcements together and a theme jumps out: both companies are selling efficiency, not just power. OpenAI is pitching Terra as better "for everyday work" and Luna as its fast, cheap option, while Musk stresses tokens and cost. This is a tell. After a year in which customers gasped at their AI bills, the labs have heard the complaint, and the frontier is shifting from raw capability toward how much that capability costs to run. Google is expected to press the same point when it releases Gemini 3.5 Pro later this month.
There is history under all of this. Musk cofounded OpenAI, left, and then sued Altman and the company in 2024, accusing them of betraying its nonprofit roots. He lost this year when a jury found he had waited too long to file. Now the rivalry plays out through product launches timed to blot out each other's headlines. For everyone else the practical takeaway is simpler and more encouraging than the drama suggests: the most capable models are becoming both more available and cheaper to use, just as governments are learning how much they want a say in when that happens.