共有:

OpenAI · GPT-5.6

The new model that waited —
gets its green light today.

Held back by the "per-customer approval" rule that landed in late June, GPT-5.6 ships today after clearing US government review. It's the first release restarted under the new process — a quiet handover from labs setting their own dates to Washington setting them.

AI Navigate Editorial2026.07.095 min read

6/30 Gov. review 7/9 today Rule live On hold GPT-5.6 BEFORE AFTER · LIVE
01

What Happened

Three weeks in the queue
under a brand-new rule

OpenAI announced GPT-5.6 in mid-June. Late that same month, the US introduced a new per-customer approval rule that requires certain high-capability models to clear a country- and company-by-company review before they can be sold. GPT-5.6 was the first frontier model caught by the filter — visible in the cloud console, but unbuyable until July 8.

On July 9, the government lifted the hold on GPT-5.6, making it the first major model to complete a full "paused → cleared" cycle under the new regime. It's a quiet but definite break from the era when labs set their own launch calendars.

Through JuneFrom July
Labs picked the launch dateRelease requires government sign-off
Same access for every customerApprovals decided customer by customer
Worldwide in weeksRegion- and firm-by-firm rollout

02

By The Numbers

Reading today's release

First
restart under the new rule
3 wks
actual review window
7/9
US government cleared
03

What To Do

How buyers should read it

The impact lands squarely on teams that were freezing model selection until approval. For everyday users, almost nothing changes.

01

Unfreeze production cutovers

Pipelines running interim on GPT-5.5 can be flipped back to 5.6 from the July 9 weekly release. API keys and rate limits inherit from your existing 5.5 quota.

02

Add "regulatory clearance" to procurement

Future point releases (5.7 and beyond) go through the same queue. Roadmaps should now carry an explicit "pending clearance" line item.

03

Personal use is unaffected

Consumer ChatGPT — web and mobile — was never in scope. Plus-and-above users see 5.6 in the model picker roll out over the course of July 9.

04

Frontier

A hinge in "who picks the launch date"

Frontier-model launches used to be governed by the labs' own calendars — mostly the competitor cycle and their earnings cycle. Today's event visibly adds a third variable: Washington's calendar. National security concerns aside, that means a same-tier rival can sometimes ship first during the review window, an outcome nobody planned for.

Today's restart is welcome, but the pattern shift is not going to reverse. From the next quarter onward, model selection needs to bake in "ship dates will move" as a working assumption.

Source: openai.com · AI Navigate — Daily Update · 2026.07.09