OpenAI's GPT-5.6 Launch Curtailed by U.S. Government Oversight, Anthropic Models Pulled Amidst Export Controls

OpenAI has initiated a restricted rollout of its newest large language model suite, GPT-5.6, limiting initial access to a select group of trusted partners. This move follows an explicit request from the U.S. government, which is dictating customer access on a case-by-case basis during this preview period.

The GPT-5.6 suite encompasses three distinct models: Sol, the flagship offering; Terra, a balanced, medium-tier option; and Luna, designed for faster, lower-cost applications. OpenAI has indicated that while it is cooperating with the administration, it views this degree of government control over model access as an anomaly, not a long-term precedent, and anticipates broader availability of GPT-5.6 in the coming weeks. The company is actively collaborating with the administration to establish a new executive order framework focused on cybersecurity and to streamline future model releases.

Advanced Capabilities and Security Posture

GPT-5.6 Sol distinguishes itself with significantly enhanced agentic capabilities across critical domains such as coding, biology, and cybersecurity. It introduces a 'max' reasoning effort mode and an 'ultra' mode that leverages coordinated subagents for complex problem-solving. Furthermore, Sol incorporates OpenAI's most robust security stack to date, engineered for formidable resilience against adversarial attacks and optimized for defensive cybersecurity operations. To validate this, OpenAI committed approximately 700,000 A100e GPU hours to automated red-teaming efforts for Sol, complemented by extensive third-party testing. Pricing for Sol is set at $5 per million input tokens and $30 per million output tokens. Terra is offered at half the price of Sol, while Luna is positioned as the most economical option at $1 per million input tokens and $6 per million output tokens.

Escalating Government Intervention and Precedents

This controlled release for OpenAI arrives amidst a broader governmental push for increased oversight of advanced AI models. The Trump administration, which previously favored a hands-off approach to AI development, recently signed an executive order in June 2026, mandating that certain AI companies voluntarily submit their most advanced models for government review up to 30 days prior to public release.

This regulatory shift has been underscored by more stringent actions, notably affecting Anthropic. While some reports suggested Anthropic voluntarily kept its most powerful AI models, Fable (Fable 5) and Mythos (Mythos 5), under wraps, multiple sources confirm the U.S. government issued a direct order, an ultimatum, and an export control directive. This directive specifically prohibited the export of Fable and Mythos outside the United States and restricted access for foreign nationals within the country. Following this mandate, Anthropic entirely withdrew both models from availability.

The government's intervention with Anthropic was partly triggered by the company reportedly granting access to Mythos to a South Korean telecom suspected of ties to China. Additionally, Amazon CEO Andy Jassy reportedly alerted the administration after Amazon researchers claimed to have circumvented Fable 5's safeguards, though Anthropic disputes the characterization of a 'jailbreak.' Anthropic's Mythos model, launched in April, had been accessible to approximately 150 vetted companies and government organizations prior to the ban.

Dean Ball, a former White House AI adviser poised to join OpenAI, has voiced criticism of the recent executive order, asserting that it effectively establishes a de facto involuntary licensing regime for frontier AI development, raising questions about the future landscape of AI innovation and deployment.