OpenAI Previews GPT-5.6 With Sol, Terra, and Luna: Tiered Models, New Reasoning Modes, Limited Access
OpenAI has begun a restricted preview of GPT-5.6, its next-generation mannequin collection. The lineup splits into three named tiers: Sol, Terra, and Luna. Sol is the flagship. Terra targets on a regular basis manufacturing work. Luna is the quick, low-cost choice.
OpenAI is beginning with a small group of trusted companions by the API and Codex. According to OpenAI put up, they shared the fashions and plans with the U.S. authorities first. Broader entry in ChatGPT, Codex, and the API is deliberate within the coming weeks.
The change is generally structural. GPT-5.6 introduces tiered fashions, two new reasoning modes, and a heavier security stack.
What is GPT-5.6?
GPT-5.6 is a household, not a single mannequin. OpenAI additionally modified the way it names releases. The quantity now marks the technology. The names mark sturdy functionality tiers.
Each tier can advance by itself schedule. That provides builders a clearer alternative throughout intelligence, velocity, and value.
OpenAI calls Sol its strongest mannequin but. It cites beneficial properties in coding, biology, and cybersecurity. Terra matches GPT-5.5 efficiency whereas costing roughly half as a lot. Luna brings sturdy functionality at OpenAI’s lowest worth.
New Reasoning Modes: max and extremely
GPT-5.6 provides two reasoning controls. The first is a brand new max reasoning effort. It provides Sol probably the most time to cause deeply.
The second is extremely mode. Instead of 1 mannequin working alone, extremely leverages subagents. These subagents break up complicated work to speed up it.
Think of it this fashion. The max setting deepens a single chain of reasoning. The extremely mode coordinates a number of employees on one process. Both commerce latency and value for accuracy on long-horizon issues.
Interactive Explainer