OpenAI on Thursday released its GPT-5.6 model family to the public, ending a period of government-mandated limited preview that had delayed the rollout since June. The three-tier series—flagship Sol, mid-range Terra, and low-cost Luna—powers a new ChatGPT Work agent designed to execute multi-step office tasks across applications, from generating documents and spreadsheets to building web apps. The launch marks a dual inflection point: the first large-scale deployment of frontier models after direct US government safety intervention, and a strategic pivot by the industry’s leading labs toward autonomous workplace agents rather than conversational chatbots alone.

The release followed what chief executive Sam Altman described as a collaborative back-and-forth with Trump administration officials, during which OpenAI made “many changes” to the models based on government red-teaming. Washington had initially asked the company to restrict early access to vetted US-based partners, citing cybersecurity concerns over the models’ code-vulnerability detection capabilities. A similar process unfolded at rival Anthropic, whose Mythos 5 and Fable 5 models were briefly subjected to export controls before restrictions were lifted in late June. Viewed from Washington, the episode revealed a regulatory apparatus relying on ad hoc export-control tools rather than a standing framework, a gap Microsoft president Brad Smith publicly criticised as “regulation without transparent or complete rules.”

The government’s interventions have had a measurable effect on the competitive landscape. In European AI circles, the temporary withdrawal of Anthropic’s models and the controlled release of GPT-5.6 accelerated interest in open-weight alternatives that cannot be remotely disabled. Chinese lab Zhipu AI’s GLM-5.2, released in early June and freely downloadable, gained traction alongside DeepSeek, which now leads in token volume on the multi-model platform OpenRouter. Analysts in Paris note that the French startup Mistral remains the only Western champion of the open approach, while some observers caution that Beijing may itself restrict open models if capabilities are deemed too risky. Enterprise customers, meanwhile, are scrutinising AI spending more sharply, with Altman acknowledging that every enterprise is now “thinking about spend and the value they’re getting.”

Both OpenAI and Anthropic have confidentially filed for initial public offerings, targeting valuations near one trillion dollars, though market conditions and regulatory uncertainty could push timelines beyond the autumn. The immediate next milestone is the development of formal criteria under a White House executive order to determine which AI models will face future safety restrictions, a process that will shape the speed and openness of the next generation of releases.