OpenAI unveils GPT-5.6 with three tiers — Sol, Terra, and Luna — and reasoning modes

OpenAI has shown a preview of GPT-5.6 — a new family of models with a three-tier architecture and expanded reasoning modes. Access is currently limited to selected developers and junior-level engineers testing the models on real tasks.

Sol, Terra, Luna: Three Tiers Instead of One Model

The GPT-5.6 family includes three models named Sol, Terra, and Luna — each with its own power profile and, presumably, token cost. This approach has long been proven by competitors: Google has the Gemini Flash, Pro, Ultra lineup; Anthropic has Claude Haiku, Sonnet, Opus. OpenAI is taking the same path, formalizing gradations within the GPT-5 generation. The logic is clear: there's no need to use the top-tier model where a lighter version will suffice. This reduces API costs, speeds up output where latency is critical, and makes the technology more accessible to small teams with limited budgets. Companies can build multi-tier systems: Sol — for quick filtering and simple answers, Luna — for complex analysis.

• Sol — basic tier, focused on speed and low token cost
• Terra — mid tier, balance of performance and quality for most tasks
• Luna — advanced tier, maximum capabilities and deep reasoning

OpenAI did not disclose exact specifications and prices at the time of announcement. The names themselves — Sun, Earth, Moon — fit the logic of increasing complexity: from the most accessible to the most powerful model in the family.

Reasoning Modes: max and ultra

GPT-5.6 introduces two new reasoning modes on top of standard output. This continues the architectural approach that OpenAI has been refining since o1: before delivering an answer, the model spends additional computational resources on "thinking" — building a chain of intermediate steps, checking logic, and revising intermediate conclusions if necessary.

The max mode is presumably intended for most complex tasks: code writing and debugging, structured analysis, multi-step questions with dependencies.

The ultra mode goes further — for extremely difficult requests: architecture design, scientific calculations, long agent scenarios where an error in an intermediate step nullifies the entire result.

An important nuance: deeper reasoning means more time for an answer and higher request cost. Engineers will need to decide when the ultra mode is truly justified, and when standard output is sufficient.

Limited Access: What You Need to Know

For now, GPT-5.6 is available only under a limited preview. OpenAI traditionally launches new architectures in stages: first internal tests and red team, then select partners and enterprise clients, and only then — wide release via API and ChatGPT.

The company has not clarified exactly when the models will be made publicly available. Early users get the opportunity to adapt their products to the new architecture before mass release and develop skills working with it ahead of competitors. If you want to get into the first wave — watch OpenAI's official API waitlist.

What This Means

GPT-5.6 reinforces the trend that has swept the entire industry: leading AI companies no longer release "one model for everything," but instead build lineups with clear price and power entry points. New reasoning modes are the next step in the race for accuracy on complex tasks. For developers, this means more flexibility and control over costs; for business, it's the ability to more accurately select a tool for a specific scenario without overpaying for excessive power.