OpenAI said on Friday that it will restrict initial access to its newest GPT‑5.6 family to a “small group of trusted partners” at the request of the U.S. government. The company stated that the preview is limited to partners “whose participation has been shared with the government.”
What’s in the GPT‑5.6 lineup?
The next‑generation GPT‑5.6 family includes three models:
- Sol — the flagship and the most powerful model in the series.
- Terra — a more balanced model for everyday use.
- Luna — a faster, lower‑cost option.
OpenAI says GPT‑5.6 Sol brings improved agentic capabilities in coding, biology, and cybersecurity. Sol introduces a “max” reasoning effort mode and an “ultra” mode that uses coordinated subagents to tackle highly complex tasks.
The government request and broader context
The announcement comes amid increasing U.S. government pressure on AI companies to limit public access to their most advanced systems. After Anthropic released its most powerful public model, Fable 5, the administration ordered Anthropic to remove access for any foreign national, and Anthropic ultimately took the model down entirely.
Dean Ball, a former White House AI adviser and soon‑to‑be OpenAI employee, has argued that the recent executive order — which asks certain AI firms to voluntarily submit their most advanced models for government review up to 30 days before release — has effectively created a de facto involuntary licensing regime for frontier AI. Ball warns that, without clearly defined safety standards, this regime could cause open‑ended launch delays, potentially disadvantaging the U.S. in the AI race and putting at risk billions of dollars in AI infrastructure investments.
OpenAI’s position and next steps
While OpenAI complied with the administration’s request this time, the company made clear it does not view this arrangement as desirable in the long term. In a Friday blog post, OpenAI said: “We don’t believe this kind of government access process should become the long‑term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.”
OpenAI described the preview as a “short‑term step” intended to lead to broader availability of GPT‑5.6 in the coming weeks, while the company and the administration work on a new executive order framework on cybersecurity and a “repeatable process for future model releases.”
Safety, usability, and pricing
To address safety concerns, OpenAI says Sol includes its most robust security stack yet: hardened against adversarial attacks and intentionally optimized to favor defensive cybersecurity work over offensive exploits. The company emphasizes that safety guardrails are built into the core model behavior rather than relying solely on an external filter, seeking to avoid problems that affected Anthropic’s Fable 5 — for instance, invisible downrouting to older models when high‑risk topics were detected.
Although GPT‑5.6 models are initially available only to a select group of partners, OpenAI plans to roll them out more broadly soon to users of ChatGPT, Codex, and the API.
Pricing for the three sizes is tiered: Sol is priced at $5 per million input tokens and $30 per million output tokens; Terra is half that cost; and Luna is $1 and $6, respectively. OpenAI also says it has improved prompt caching to make repeated prompts cheaper and more predictable.
