OpenAI’s GPT-5.6 Sol Launches Under Tight Government Control

OpenAI rolled out its most powerful AI model yet, GPT-5.6 Sol, but only to 20 government-approved partners. This restricted preview follows a direct request from the US government to stagger access carefully.

This marks the first time an American AI company has launched a frontier model under a government-managed access list. Each partner’s approval passed through a customer-by-customer vetting process. OpenAI’s CEO Sam Altman confirmed the government reviewed access during the preview and hopes for a broader release in a few weeks if all goes well.

Sol leads a new three-tier series that includes Terra and Luna, with tiered pricing: Sol at $30 per million output tokens, Terra at half that, and Luna at $6. OpenAI also improved prompt caching to reduce costs. The model excels in coding, biology, and cybersecurity, outperforming Anthropic’s Claude Mythos 5 in coding workflows while using only a third of the output tokens.

OpenAI introduced two new modes for Sol: a “max reasoning effort” mode for peak problem-solving and an upcoming “ultra” mode that splits tasks among coordinated sub-agents. Sol’s security stack is robust, hardened against adversarial attacks, and optimized for cybersecurity defense. Safety guardrails are built directly into the model, avoiding sole reliance on external filters.

Sol is also available through Amazon Bedrock. OpenAI plans to make all three tiers generally available to ChatGPT, Codex, and API users soon. Still, the company insists this government-managed access should not become a permanent approach. “We’ve made clear to the U.S. government that this is not our preferred long term model,” Altman said, aiming for a sustainable release strategy in collaboration with the industry.

The US government’s cautious rollout likely stems from a 2023 Trump administration executive order calling for voluntary review of new AI models. This context includes Anthropic’s recent troubles, where its Claude model was pulled offline after a federal shutdown order linked to security concerns. Anthropic accused China of unauthorized access through nearly 25,000 fake accounts generating over 28.8 million interactions with Claude.

A White House spokesperson emphasized ongoing collaboration with frontier AI labs, stating, “The White House continues to collaborate with frontier AI labs to develop shared approaches for addressing the challenges of scaling this technology.”

Meanwhile, Microsoft quietly extended free Windows 10 ESU support to October 12, 2027, without any formal announcement. This update revises the post-end-of-support period, originally set to end October 14, 2025. Enterprises pay $427 per device for ESU, but this move is unlikely to grab headlines amid the AI frenzy.