OpenAI Launches GPT-5.6 Family With Three Model Tiers: Sol, Terra and Luna

OpenAI has announced the GPT-5.6 family of models, rolling them out across ChatGPT, its Codex coding tool, and the OpenAI API. The new family comprises three variants — Sol, Terra, and Luna — each pitched at a different point on the cost-versus-capability spectrum.

The launch marks one of the company’s largest model releases to date, covering professional coding, scientific research, knowledge work, and cybersecurity tasks. OpenAI describes GPT-5.6 as a “frontier” large language model family built for enterprise and professional use at scale.

Three Models, Three Price Points

Sol is the flagship. It’s designed for the most demanding tasks — complex reasoning, advanced coding, and so-called “agentic” workflows where the AI carries out multi-step tasks with minimal human input.

Terra sits in the middle. OpenAI says it performs broadly on a par with the previous GPT-5.5 generation but at a lower price, making it the practical choice for businesses that don’t need Sol’s full horsepower.

Luna is the fastest and cheapest of the three. It’s built for high-volume applications where speed matters more than depth — think customer-facing chatbots, rapid content generation, or internal query tools fielding thousands of requests a day.

API pricing, quoted per one million tokens, breaks down as follows: Sol costs $5 for input and $30 for output — around £4 and £24 respectively. Terra runs at $2.50 input and $15 output — around £2 and £12. Luna comes in at $1 input and $6 output — around 80p and £4.70.

Who Gets Access to What

Not everyone sees all three models. ChatGPT Plus, Pro, Business, and Enterprise subscribers can choose between Sol, Terra, and Luna within ChatGPT Work and Codex. Free and Go users are generally limited to Terra in certain modes, with Sol and Luna sitting behind the paywall.

The rollout is gradual. OpenAI has been staging access — starting with a limited preview for trusted partners via the API and Codex — before opening up more broadly. The company says full availability is coming within a short launch window, though the exact timeline depends on region, plan type, and workspace settings.

GPT-Live-1, a new voice model, launched alongside GPT-5.6 and is set as the default voice option for paying ChatGPT users. The two product lines are designed to work together across OpenAI’s platform.

Safety and Cybersecurity Concerns

OpenAI’s own Preparedness Framework classifies all three GPT-5.6 models — Sol, Terra, and Luna — as “High capability” in cybersecurity and biological/chemical risk domains. But the company says none of them reach the “Critical” risk level.

That classification matters. OpenAI’s safety analysis found that the models can help identify software vulnerabilities and fragments of potential exploits. But they cannot, the company says, autonomously carry out end-to-end attacks on hardened systems or produce fully functional zero-day exploits without a human in the loop.

The US government requested a limited preview period before the general release, during which access was restricted while authorities reviewed the models’ capabilities. Those restrictions have since been relaxed as the wider rollout proceeds.

Sam Altman, OpenAI’s chief executive, said the company had worked closely with government partners during the preview phase, describing the staged approach as part of OpenAI’s commitment to deploying powerful models responsibly.

Prompt Caching and Technical Changes

GPT-5.6 also brings updated prompt caching — a technical feature that lets developers store parts of a conversation or query to avoid reprocessing them on every request. The minimum cache life is 30 minutes. Cache writes are billed at 1.25 times the standard input rate; cache reads get a 90 per cent discount. For developers running high-volume applications, that discount can add up quickly.

The tiered model structure mirrors a broader shift across the AI industry. Competitors including Google with Gemini 2.5 and Anthropic with Claude have adopted similar approaches — offering a range of models at different price points rather than a single flagship. OpenAI is following that pattern with GPT-5.6.

Industry Reaction

Responses to the launch have been mixed, as they tend to be with major AI releases. Enterprise users and developers have broadly welcomed the flexibility — three models, clear pricing, and a choice of capability levels. Cybersecurity specialists, however, have noted that more capable models carry inherent dual-use risk, even if OpenAI’s own testing places the models below the highest danger threshold.

Civil society groups have raised longer-standing concerns: the concentration of powerful AI capabilities inside a handful of large US-based companies, questions about training data transparency, and the potential for job displacement in knowledge-work roles. OpenAI has not addressed those concerns directly in its launch materials.

Academic researchers see practical value in GPT-5.6 for literature review, coding assistance, and data analysis, while flagging that hallucination — the tendency of AI models to generate plausible-sounding but incorrect information — remains a live issue across the GPT-5.x generation.

What This Means for Kent Residents

Kent residents on free ChatGPT plans will largely access GPT-5.6 through Terra, the mid-tier model, with Sol and Luna sitting behind paid subscriptions. Businesses and organisations in Kent — including local councils, NHS Kent and Medway ICB, and software companies — that hold Plus, Pro, Business, or Enterprise accounts can now choose between all three models for tasks such as document drafting, code review, and data analysis, though any public-sector use will need to be weighed against UK GDPR obligations and national guidance on AI in public services. For Kent-based SMEs and developers, the lower-cost Luna tier makes high-volume AI integration more affordable than previous generations allowed.