OpenAI Unveils GPT-5.6 Family — but Locks It to a Handful of Partners
OpenAI has pulled the wraps off its next generation of models, a GPT-5.6 family split into three tiers — Sol, Terra and Luna — but for now access is being rationed to roughly 20 organisations rather than the wider public.
Sol is the flagship, pitched at advanced coding, scientific reasoning, long-horizon planning and the kind of multi-step 'agentic' workflows the industry is racing to make reliable. Terra is billed as the everyday workhorse, matching the outgoing GPT-5.5 while costing roughly half as much, and Luna is the fast, cheap option aimed at high-volume tasks. A new 'ultra' mode promises deeper reasoning on the hardest problems.
Pricing, quoted per million tokens, underscores the tiering: Sol at 5 dollars input and 30 dollars output, Terra at 2.50 and 15, and Luna at 1 and 6. OpenAI is also promising tougher safety protections and more predictable prompt caching as it prepares for broader availability.
The restrained rollout is as notable as the technology. OpenAI says it shared the models and its release plans with the U.S. government before granting even limited access, a nod to mounting concern over advanced systems' capabilities in sensitive domains such as biology and cybersecurity. A general release, the company says, is planned for 'the coming weeks'.
OpenAI is separately launching Sol on specialised Cerebras hardware at speeds reported to reach hundreds of tokens per second, initially for select customers as capacity expands. Taken together, the GPT-5.6 debut signals both how fast frontier capability is advancing and how cautiously — and commercially — its makers now intend to release it.
