OpenAI Launches GPT-5.6 Family of Three AI Models in US Government-Coordinated Limited Preview

OpenAI announced on 26 June 2026 a new generation of AI models under the GPT-5.6 family name, comprising three distinct tiers: GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna. The launch marks a shift away from single-flagship releases toward a structured family of models differentiated by capability and cost — but for now, almost nobody can use them.

The preview is running “at the request of the US government”, according to OpenAI, and access is restricted to a small group of trusted partners and organisations with an existing OpenAI account representative. There is no public waitlist, no self-service sign-up, and the models are not yet available inside ChatGPT. Access is limited to OpenAI’s API and Codex, the company’s developer-focused coding environment.

OpenAI has said it believes in broad access and intends to make all three models “generally available in the coming weeks” across ChatGPT, Codex and the API. But for the moment, the preview is, in OpenAI’s own words, “not a broad self-service programme.”

Three Models, Three Price Points

The GPT-5.6 family is built around a tiered structure. Sol sits at the top as the flagship model, designed for complex reasoning and demanding tasks. Terra occupies the middle ground — a balanced option that OpenAI describes as delivering performance comparable to the previous GPT-5.5 flagship at roughly half the cost. Luna is the fastest and cheapest of the three, aimed at high-volume workloads where speed and economy matter more than raw capability.

Pricing during the preview, cited by DataCamp from OpenAI’s published figures, runs as follows per one million tokens: Sol is priced at around $5 (about £4) for input and $30 (about £24) for output; Terra at $2.50 (about £2) input and $15 (about £12) output; Luna at $1 (about 80p) input and $6 (about £4.75) output. Those figures apply to the API preview and may change at general availability.

The naming convention itself — Sol, Terra, Luna — is new for OpenAI, and represents a deliberate move to brand capability tiers rather than iterate on a single model number.

Safety Testing and the Government Connection

OpenAI published a preview system card alongside the announcement, describing enhanced safety measures for the GPT-5.6 generation. According to the company, these include more rigorous red-teaming processes and expanded automated testing, reflecting what OpenAI frames as its approach to deploying frontier AI responsibly.

The explicit involvement of the US government in shaping the preview’s structure is notable. OpenAI has stated directly that the limited rollout was coordinated at governmental request, suggesting a degree of oversight built into the earliest phase of deployment. Early use cases are focused on software development, cybersecurity, scientific research and enterprise workloads — not consumer chat applications.

Community reports and media commentary suggest the trusted partner group numbers around 20 organisations or roughly 100 individuals, though OpenAI has not confirmed those figures officially, and they should be treated as unverified.

Who Gets Left Out — and for How Long?

The restrictions have drawn criticism from developers and smaller firms who argue that government-coordinated gated access concentrates powerful capabilities among a narrow set of corporate and governmental actors. Some commentators have raised concerns about transparency and whether the preview structure disadvantages international users and smaller organisations that lack direct relationships with OpenAI’s enterprise sales teams.

Developers, in particular, have expressed frustration at the absence of a public evaluation process and limited technical documentation during the preview window. Many are calling for clearer timelines and more open access to benchmarking.

OpenAI’s response, broadly, is that the preview structure exists for risk management and regulatory engagement, and that general availability is coming soon. The company has not yet specified a precise date.

Some community sources have reported that GPT-5.6 is not currently available in the EU or the UK during the preview phase, citing regulatory and policy considerations as a possible factor in the rollout sequence. OpenAI has not formally confirmed geographic exclusions in its published documentation, so that status remains unverified.

Sam Altman, OpenAI’s chief executive, has consistently stated the company’s intention to make its most capable models widely accessible. He has not, at the time of writing, made a specific public statement about the GPT-5.6 preview timeline for non-US markets.

What Comes Next

OpenAI has signalled that broader availability may be rolled out on a country-by-country basis. The company has not published a specific schedule for when GPT-5.6 Sol, Terra or Luna will appear inside ChatGPT for standard subscribers, nor confirmed when API access will open to organisations outside the current trusted partner group.

The key unanswered questions are: when exactly does general availability begin; which countries are included from day one; and whether the tiered pricing confirmed for the preview will carry over unchanged to the public launch.

What This Means for Kent Residents

During the current preview phase, individuals and organisations in Kent — including local businesses, universities, colleges and council digital services — cannot access GPT-5.6 Sol, Terra or Luna, as participation is restricted to a small group of pre-selected partners. Anyone currently using OpenAI tools through a standard account will continue on existing models such as GPT-5.5 until general availability is confirmed. If unverified reports of a UK exclusion during the early rollout prove accurate, Kent users and UK organisations more broadly may face a short additional wait compared with their US counterparts before the new model family becomes accessible.