OpenAI launches the GPT-5.6 family in three variants: Luna, Terra, and Sol

On July 9 OpenAI released GPT-5.6, following a limited preview for selected partners that began on June 26. The structural novelty: not a single model but a family of three, with astronomical names and distinct roles. Luna is the fastest and cheapest, Terra the middle tier (claimed to match its predecessor GPT-5.5 at half the cost), and Sol the flagship for reasoning, code, and scientific research.

The logic is that of a tiered price list: instead of paying for the best model for everything, the user or developer picks the level the task actually needs. Translating an email does not require the same power as a security audit of a codebase, and the prices reflect that: from Luna's 1/6 dollars per million tokens up to Sol's 5/30.

Alongside the models comes ChatGPT Work, a tool built for structured office work with Sol as its agentic engine. It is the clearest signal of the commercial direction: less general-purpose chatbot, more platform that executes complete work tasks.

An unusual emphasis is cybersecurity: OpenAI presents Sol as its strongest model ever for threat analysis, code review, and patching. A double-edged sword acknowledged openly, reigniting the debate over whether these tools help defenders or attackers more.