OpenAI has begun a limited preview of GPT-5.6 Sol, its most capable model yet, alongside the cheaper Terra and Luna tiers, with the family rolling out to a handful of partners first this week.
The rollout marks OpenAI’s biggest model update since GPT-5.5 and lands as rivals push their own frontier systems into the wild.
The three tiers are pitched at different jobs and budgets, with the company planning wider general availability in the coming weeks, as reported by VentureBeat.
What GPT-5.6 Sol, Terra and Luna do
Sol is built for the hardest problems, such as complex coding and security research.
Terra targets high-volume business tasks like customer support, internal tools and document analysis, while Luna handles faster, lower-cost everyday work such as summarisation, drafting and routine automation across the board.
The headline feature is a new max reasoning setting for Sol, aimed at problems that need extended deliberation. OpenAI is also introducing an ultra mode, which brings in subagents that can split up and accelerate complex projects, chaining steps together across the whole model family.
For anyone who has tracked OpenAI’s naming, the jump to celestial codenames is new. Sol, Terra and Luna replace the old numbered sprawl with a sun, earth and moon hierarchy that maps neatly onto power and price, with the brightest and most expensive model sitting at the very top.
How much GPT-5.6 costs
Pricing splits cleanly across the three tiers. Sol sits at the top at $5 per million input tokens and $30 per million output tokens, the same as GPT-5.5.
Terra runs at $2.50 and $15, while Luna is the cheapest at $1 and $6 per million tokens in and out.
Sol also launches with what OpenAI calls its most robust safety stack to date, with strengthened protections for higher-risk activity, sensitive cyber requests and repeated misuse.
The guardrails reflect how capable the flagship has become at coding and security work that could be turned to harm.
Who can use GPT-5.6 Sol right now
Access is tight for the moment. The models are going first to a narrow set of roughly 20 organisations, after OpenAI shared the systems and its release plans with the United States government.
The gatekeeping points to growing scrutiny of frontier models before they reach the wider public.
The performance pitch leans on stronger coding, scientific reasoning, long-horizon planning and agentic workflows. In plain terms, OpenAI wants Sol to chew through tasks that take humans hours, chaining the steps together without a person nudging it along at every single turn.
OpenAI says it plans to make Sol, Terra and Luna generally available in the coming weeks.
For developers and businesses, the open questions now are how the new max reasoning and ultra modes hold up under real workloads, and whether the cheaper Terra and Luna tiers undercut rivals hard enough on cost.
