Commerce clears GPT-5.6 for Thursday launch as Anthropic extends Fable 5 reprieve to July 12

Both frontier labs' flagships spent Tuesday navigating the Trump administration's case-by-case AI oversight regime.

The U.S. Department of Commerce cleared OpenAI on Tuesday to run a broad global launch of GPT-5.6 on Thursday, ending a limited-preview regime that had confined the three-tier Sol/Terra/Luna family to vetted partners since last month. Hours later, Anthropic extended free Fable 5 access for paid subscribers through 11:59:59 PM PT on July 12, walking back a cutoff that had triggered user backlash. Two flagship rollouts, one afternoon, both routed through the same case-by-case approval machinery the Trump administration has built out of Commerce.

The mechanics are worth pausing on. Per Axios, the Center for AI Standards and Innovation ran its own evaluations on GPT-5.6, and OpenAI kept technical staff on the ground in Washington to field follow-up questions from officials. Bloomberg quotes an OpenAI spokesperson saying the preview is expanding worldwide “with a green light from US government leadership.” Neither OpenAI, the White House, nor Commerce returned Reuters’ requests for comment.

OpenAI’s own posture is more ambivalent than the spokesperson line suggests. The GPT-5.6 system card claims Sol sets a new state of the art on Terminal-Bench 2.1 and matches Anthropic’s Mythos Preview on ExploitBench using roughly one-third the output tokens, alongside a “max reasoning effort” setting and an “ultra mode” that dispatches subagents. Buried in the same document: “We don’t believe this kind of government access process should become the long-term default.”

Anthropic’s Tuesday is the mirror image. Commerce banned foreign access to Mythos and Fable in June; the Fable restriction was lifted only last week, and now a domestic pricing cliff has been softened, with the 50% weekly-usage cap on Fable 5 during the reprieve intact.

The pattern is legible enough. Frontier deployment now clears through Washington, one exception at a time, and the labs are learning to write system cards that thank the regulator on page one and object to the regime on page forty.