Enterprise

OpenAI debuts GPT-5.6 Sol behind a government-vetted gate, warns the model is its strongest yet

The Trump administration asked OpenAI to limit Friday's launch of Sol, Terra, and Luna to a small set of vetted partners — a constraint the company says shouldn't become the default.

By Asher Glick · Enterprise software desk · 2026-06-29 · 3 min read

Photo: Unsplash / Andy Feliciotti — The dome of the U.S. Capitol building in Washington, D.C.

OpenAI on Friday introduced GPT-5.6 Sol alongside two siblings, Terra and Luna, and disclosed in the same announcement that the Trump administration asked it to restrict the launch to a small group of trusted partners whose identities were shared with the U.S. government before any wider release. The company complied and then, in unusually direct language, said this kind of pre-clearance process shouldn’t become the long-term default because it keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.

The framing matters. OpenAI is calling the limited preview a short-term arrangement tied to a forthcoming cyber Executive Order framework, signaling that a repeatable government-access pipeline for frontier model releases is now being negotiated in real time. Reuters reported the administration’s rationale: early access to identify threats, from cyberattacks to military misuse, before frontier tools are widely deployed.

Anthropic offers the immediate precedent. The company recently disabled two of its latest models to comply with a Trump administration export control directive, and Claude Mythos 5 was effectively banned this month. What looked a few weeks ago like a one-off has resolved into a pattern.

The product itself is positioned aggressively. Sol introduces a “max” reasoning effort mode and an “ultra” mode that uses coordinated subagents to solve highly complex tasks, and OpenAI claims it edges Claude Mythos 5 on coding workflows. Pricing slots Sol at $5 per million input tokens and $30 per million output tokens; Terra costs half that, at $2.50 and $15; Luna lands at $1 and $6.

OpenAI’s own system card classifies all three as High capability in both Cybersecurity and Biological and Chemical risk under its Preparedness Framework, while none reach the High threshold in AI Self-Improvement. In cybersecurity testing the models could find vulnerabilities and pieces of exploits but couldn’t carry out autonomous, end-to-end attacks against hardened targets, and Sol doesn’t cross the “critical” cybersecurity threshold. Broader availability across ChatGPT, Codex, and the API is coming in the next few weeks.

Read structurally, Friday’s announcement is the moment frontier AI release cycles started looking like defense export controls. The lab gets to ship; the state gets first look; the public waits. OpenAI’s objection, filed alongside its own compliance, is the tell.

Sources