OpenAI has launched GPT-5.6 Sol to 20 government-approved partners in a limited preview.

      TL;DR OpenAI has launched Sol, its most advanced model, to approximately 20 partners approved by the government under Trump's AI directive. OpenAI has unveiled GPT-5.6 Sol, its leading model, granting access to about 20 government-approved partners. This release marks the first instance of an American AI company distributing a frontier model under a government-regulated access list, going beyond the voluntary pre-release review system established by Trump's AI executive order on June 2. Sol represents the most advanced model in a new three-tier lineup, which also features Terra, a mid-tier option, and Luna, focused on speed and cost. OpenAI highlighted Sol's strengths in coding, biology, and cybersecurity, introducing a new “max reasoning effort” mode that allows the model more time to tackle complex issues. The company also intends to roll out an “ultra” mode that divides tasks among several sub-agents.

      This limited preview comes at the request of the Trump administration, with the government approving partners one at a time during the preview phase, as reported by Bloomberg. OpenAI indicated in a blog post that it does not think this type of government access process should become a long-term norm but decided to take part anyway. This arrangement serves as the first practical evaluation of the executive order signed by Trump earlier this month, which encourages AI companies to voluntarily provide the government with up to 30 days of pre-release access to models with advanced cyber capabilities. The order expressly rejects mandatory licensing, but the precedent set by Anthropic lends it authority. Two weeks ago, the government instructed Anthropic to disable Fable 5 and Mythos 5 following a reported jailbreak, marking the first instance of the government forcibly taking a commercial AI model offline.

      OpenAI's choice to collaborate contrasts with Anthropic's experience. Although Anthropic complied with the shutdown order, it criticized the action as excessive, warning that it would cease all frontier model launches if such measures became widespread. In contrast, OpenAI appears to embrace voluntary compliance as a means to avert a coercive scenario while maintaining the right to challenge the principle. Sol is also accessible through Amazon Bedrock, making it the first model in the new series available on a competing cloud service. OpenAI intends to release all three tiers to the public in the coming weeks, although no specific date has been announced.

      A pressing concern remains whether government-controlled releases will set a precedent for all future frontier models. OpenAI clearly aims to prevent this and has voiced its stance publicly. However, with Anthropic's models still offline and the voluntary framework of the executive order yielding seemingly mandatory outcomes, distinguishing between cooperation and compliance is becoming increasingly difficult.

      Published June 26, 2026 - 6:25 pm UTC