Google updated its Gemini 3.5 model page on July 6, pushing Gemini 3.5 Pro into broader distribution on Vertex AI while OpenAI's GPT-5.6 Sol and Anthropic's Fable 5 sit behind what Axios calls "the first time the U.S. government has preemptively asked a U.S. AI company to restrict the launch of a model before release." One vendor is shipping. Two are being reviewed by the Office of the National Cyber Director and the Office of Science and Technology Policy. That's the review this week.

The Google page positions Gemini 3.5 Pro as the higher-capability orchestration tier above Gemini Flash, and it leans on institutional customer quotes to do the heavy lifting. Thomson Reuters CTO Joel Hron calls the model "state-of-the-art reasoning with depth and nuance," citing "measurable and significant progress in both legal reasoning and complex contract understanding." Wayfair describes "a clear step forward in handling structured business tasks that require precision and consistency." These are the reference customers a frontier vendor lines up when it wants to look boring and dependable, which is currently the most valuable posture in the market.

The contrast on the OpenAI side is sharper than the model spec. On June 26, OpenAI announced three GPT-5.6 variants, Sol, Terra, and Luna, with Sol described by CNBC as its strongest model to date, showing gains in coding, biology, and cybersecurity. OpenAI says Sol doesn't cross its internal "critical" cybersecurity risk threshold and is "better at helping users fix vulnerabilities than it is at carrying out end-to-end attacks." The models were nonetheless restricted to a "small group of trusted partners" whose participation is "shared with the government." Axios pegs that group at roughly 20, with Amazon Bedrock as one route in. Sam Altman told staff the government is "approving access customer by customer," with a broader release hoped for "a couple of weeks later."

Anthropic's Fable 5 fared worse. After release, the administration ordered Anthropic to strip access for any foreign national; Anthropic pulled the model entirely.

The framing this all sits inside is a Trump executive order asking companies to submit advanced models for federal review up to 30 days before release under a nominally voluntary protocol. Dean Ball, a former White House AI adviser and soon-to-be OpenAI employee, called the arrangement a "de facto involuntary licensing regime for frontier AI." His follow-up is the sentence that matters: "The government doesn't have clearly defined safety standards."

So the review reads like this. Gemini 3.5 Pro is on Vertex AI with reference customers. Sol is on Bedrock for about twenty accounts. Fable 5 is offline. Distribution, not capability, is the axis on which frontier AI is now being sorted, and the sorting is being done in Washington.

Gemini 3.5 Pro enters gradual rollout as GPT-5.6 Sol and Fable 5 stay gated to ~20 partners

Sources