GPT-5.6 goes public with three price tiers and the first White House model sign-off

OpenAI has released GPT-5.6 as a three-model family: Sol, Terra, and Luna. Pricing runs from $5 per million input tokens and $30 per million output tokens for Sol to $1 and $6 respectively for Luna, with Terra positioned at $2.50 and $15. Before the public launch, the models spent twelve days under review by the White House under a voluntary AI safety framework — the first structured pre-release government evaluation of a major AI model family.

GPT-5.6 Sol is available through ChatGPT, ChatGPT Work, Codex, and the OpenAI API. Microsoft has designated Sol as the preferred model for Copilot 365, meaning it becomes the default for the enterprise productivity suite used across its corporate customer base. The three-tier structure marks a departure from OpenAI’s previous pattern of releasing a flagship plus one or two derivative models: Sol, Terra, and Luna are designed from the start as a deliberate enterprise, mid-market, and cost-sensitive tier — a model line mapped to different buyer types rather than different use cases within a single buyer category.

The White House review is the more structurally novel element of this launch. Under voluntary commitments from the 2023 AI Safety Summits, major AI companies agreed to share capabilities evaluations with the government before release. GPT-5.6 is the first major model family to go through this pipeline with a disclosed timeline. What the twelve-day review evaluated — which capabilities, which thresholds, which testers — has not been made public. OpenAI’s characterization of it as a safety gate is accurate in the sense that it happened; it says less about what the gate tested or what would have caused it to stay closed.

The limits of voluntary frameworks matter here. OpenAI decided what to submit, when to submit it, and when the review period ended. The White House was a reviewer in a process the company designed — not an independent safety authority with power to delay or block the release. That arrangement does not make the evaluation meaningless: it means the credibility of the gate depends entirely on what OpenAI disclosed and how the government chose to evaluate it, neither of which is publicly verifiable. Sol, Terra, and Luna are public regardless of what the review found; the policy disclosure is that a review happened.

GPT-5.6 is the third major model release from OpenAI in 2026, following the GPT-5 family and GPT-Live. The company has not disclosed whether the White House review process will apply to all future model releases or was specific to this family. Microsoft’s decision to designate Sol as the Copilot 365 default means the model reaches enterprise users through a channel OpenAI does not control directly — Copilot’s integration into Office applications covers a set of workflows where the underlying model is largely invisible to the person using it.

Tags: OpenAI, microsoft, chatgpt, AI models, GPT-5.6, Copilot 365