OpenAI Restricts GPT-5.6 Sol, Terra, and Luna to "Trusted Partners" at U.S. Government's Request

A new era of government gatekeeping for AI

OpenAI on June 26, 2026, unveiled three new frontier AI models — GPT-5.6 Sol, Terra, and Luna — but immediately restricted access to a "small group of trusted partners" whose participation was cleared with the U.S. governmentcnbc +1. Sol is OpenAI's most powerful model to date, with gains in coding, biology, and cybersecurity; Terra is a balanced mid-tier option; and Luna is a faster, lower-cost varianttechcrunch. The rollout marks the first time the Trump administration extended its AI customer-vetting process beyond rival Anthropic, setting a precedent that rattled developers and policy observerswashingtonpost.

What the models can — and can't — do

GPT-5.6 Sol introduces a "max" reasoning-effort mode and an "ultra" mode that deploys coordinated subagents for complex taskstechcrunch. OpenAI says Sol is competitive with Anthropic's Claude Mythos 5 on coding benchmarks while using roughly a third of the output tokens, and its safety guardrails are embedded in the core model rather than applied as a separate filtertechcrunch. Sol's cybersecurity features are designed to favor defense over offense and don't cross OpenAI's "critical" risk threshold — defined as enabling "unprecedented new pathways to severe harm"cnbc. Pricing puts Sol at $5 per million input tokens and $30 per million output tokens, Terra at half those rates, and Luna at $1 and $6 respectivelytechcrunch.

A rapid reversal in Washington

The restrictions reflect a striking shift in U.S. AI policy. Trump returned to office promising deregulation and decrying Biden-era safety standards, but cybersecurity-capable frontier models have since alarmed officialswashingtonpost. His executive order this month asked developers to voluntarily submit advanced models for government review up to 30 days before release — a move critics say has hardened into a de facto licensing regimetechcrunch +1. Sam Altman wrote on X that he doesn't "like the idea of the government picking the customers," while Dean Ball, a former White House AI adviser set to join OpenAI, called the shift a move from "implausibly libertarian to increasingly draconian and opaque"washingtonpost.

Anthropic has faced a harsher version of the same pressure. After the Commerce Department placed export controls on its Mythos 5 and Fable 5 models, Anthropic pulled them entirely. A subsequent Commerce letter reviewed by The Washington Post permitted Mythos 5 to return only for roughly 100 U.S.-based companies, with the government reserving the right to revise that list at willwashingtonpost. Britain's AI Security Institute is reportedly the only non-U.S. entity with approved access to GPT-5.6washingtonpost.

OpenAI complies — but objects

OpenAI was explicit about its reluctance. "We don't believe this kind of government access process should become the long-term default," the company wrote. "It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."cnbc +1 The company called the restricted preview a "short-term step" as it works with the administration on a "repeatable process for future model releases" — with a broader rollout planned within weekscnbc.