OpenAI's GPT-5.6 goes public Thursday after US standoff

OpenAI will make its most advanced model family widely available on Thursday, capping a monthlong standoff with the US government over how powerful AI systems should be released.

The company said Tuesday night that GPT-5.6 Sol, along with the lower-cost Terra and Luna tiers, will launch publicly and that preview access is expanding globally starting now.

Initially, OpenAI restricted access to GPT-5.6 only to a few handpicked government-approved companies in mid-June following the push from the Trump administration for phased deployment due to national security risks posed by the cyber capabilities of the model.

The company explained then that it did not believe it was a sustainable model, explaining that it prevented the innovative technology from reaching the developers and enterprises that needed it.

According to Axios, the clearance comes after an additional examination conducted by the Center for AI Standards and Innovation, while OpenAI employees remain in Washington to answer any possible questions the officials have.

It is a sign of the more general change in direction towards federal examination on a case-by-case basis of the frontier models before their wide availability.

The situation is a repetition of a case where Anthropic's Fable and Mythos faced export restrictions by the Commerce Department due to possible misuses of their capabilities.

This was resolved after lengthy negotiations with the officials, although Fable and Mythos have been subjected to different timelines during the procedure. The two examples collectively form one of the clearest instances of Washington trying to negotiate access to AI technology from individual firms rather than using rigid policies.

GPT-5.6 consists of three models, Sol, the most powerful member of the family; Terra, which has been developed for everyday use and is around twice as cheap as the former per token; and Luna, the fastest and cheapest model in the family. ChatGPT itself operates on the GPT architecture, which is also used by other firms to develop AI technologies and services.