For the vast majority of the AI developer community, the most capable model OpenAI has ever built is currently a ghost. While the GPT-5.6 Sol flagship model was previewed on June 26, 2026, it is not available to the public. Instead, access is restricted to roughly 20 unnamed trusted partners, all of whom must pass through a federal vetting process.
This is not a standard beta release. In a memo to staff, Sam Altman noted that the government would be “approving access customer by customer during this preview period.” This bottleneck is the direct result of Executive Order 14409, which established a framework requiring developers to provide the federal government with up to 30 days of pre-release access to any model designated as a “covered frontier model.”
The technical architecture of Sol further complicates the access landscape. The model is deployed on Cerebras WSE-3 wafer-scale silicon, utilizing a disaggregated architecture that enables speeds of up to 750 tokens per second. Sol introduces an ultra mode, which fans tasks out to coordinating subagents, and a max mode for extra deliberation. In Terminal-Bench 2.1, these modes achieved scores of 91.9% and 88.8% respectively.
OpenAI has publicly pushed back: “We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.”
Experts have warned that the combination of undefined government safety standards and the resulting launch delays could provide a strategic advantage to Chinese AI labs. The most powerful tools are no longer just for sale — they are for the cleared.