OpenAI previews GPT-5.6 Sol — gated to ~20 partners at the US government's request

OpenAI is opening a limited preview of its GPT-5.6 family: Sol, the new flagship; Terra, a mid-tier model for everyday work; and Luna, a cheaper, faster option. Sol is pitched as the company’s most capable model to date, with stronger agentic performance in coding, biology, and cybersecurity. It adds two new reasoning tiers — ‘max,’ which lets the model think longer, and ‘ultra,’ which spins up subagents to parallelize complex tasks. Pricing for Sol holds steady at $5 per million input tokens and $30 per million output tokens, matching GPT-5.5.

The more consequential story is the rollout itself. Rather than a broad launch, the models are going to roughly 20 vetted organizations after OpenAI shared the systems and its release plan with the U.S. government — and, per multiple reports, after the White House asked OpenAI to hold back wider access. The caution tracks with the models’ dual-use profile: OpenAI says Sol and Terra can locate vulnerabilities and assemble partial exploits, but in testing could not carry out autonomous, end-to-end attacks against hardened targets. The company says it paired the release with its most robust safety stack yet.

The significance is less about benchmarks than precedent. This is one of the first visible cases of government pressure throttling a frontier model launch, with offensive-cyber and bio capability as the explicit rationale. OpenAI says general availability for Sol, Terra, and Luna will follow in the coming weeks, making this preview a test case for how frontier releases get governed when capabilities start to brush against national-security thresholds.