GPT-5.6, Explained: What Actually Changes in ChatGPT With Sol, Terra and Luna

Four days after OpenAI launched GPT-5.6, it is worth separating the noise from what has actually changed. The short answer: for free ChatGPT users, almost nothing for now; for paid plans, a new reasoning engine; for developers, three new price points that reshuffle the market. The public release reached ChatGPT, Codex and the API on July 9, after two weeks of restricted rollout with the U.S. government in the loop.

Three names, one family

GPT-5.6 is not one model but three tiers of the same generation, named after celestial bodies. Sol is the top of the range: $5 per million input tokens and $30 per million output. Terra is the mid-tier for everyday work ($2.50/$15), and Luna the budget option ($1/$6). According to OpenAI, Sol scores 80 on Artificial Analysis' coding agent index — "2.8 points above Fable 5," Anthropic's flagship — while using "less than half the output tokens," with 54% better token efficiency in coding; Terra would perform "just above Fable 5" and Luna "outperforms Opus 4.8." These are claims by the company and the indexes it cites: thorough independent comparison is still under way. One cross-cutting novelty: OpenAI presents the family as its "strongest cybersecurity model yet," aimed at threat modeling and code review and patching.

What changes (and what doesn't) in ChatGPT

Here it pays to go slowly, because the name misleads. According to OpenAI's official documentation, GPT-5.5 Instant remains ChatGPT's default model for quick everyday answers: anyone opening the app today will not notice an engine swap. The change lives in the reasoning options: on Plus, Pro, Business and Enterprise plans, medium-and-above effort levels now run on Sol, and Pro and Enterprise subscribers also get "Sol Pro," the variant for the hardest tasks. Terra and Luna, by contrast, do not appear in ChatGPT's model picker: they live in the API, for people building applications.

The other product-shaped novelty is ChatGPT Work, an assistant for enterprise teams — desktop, web and mobile — aimed at daily clerical work: drafting documents, spreadsheets and presentations. And the rollout was unusually coordinated: on July 9 itself, GitHub switched on all three models in Copilot (Sol for the higher plans; off by default on business accounts until an administrator enables them).

A launch with the government inside

The least usual part of this story is not technical. GPT-5.6 first appeared on June 26 in a phase limited to "trusted partners," a restriction OpenAI adopted at the request of the U.S. government. Before the wide rollout, the Commerce Department's Center for AI Standards and Innovation ran additional tests on the model, and the company sent technical experts to Washington to answer questions as they arose. OpenAI has kept its distance from the precedent: "We don't believe this kind of government access process should become the long-term default," the company said. It is the second episode this summer in which a frontier launch went through prior government review, after the Claude Fable 5 case in June — a new pattern that deserves its own follow-up.

For the reader, two checks are worth more than any headline: the model picker on your ChatGPT plan — where you will see whether reasoning already runs on Sol — and, if you write code, the API cost calculator, because this generation's real battle is fought as much on price per token as on benchmarks. What GPT-5.6 confirms is that the 2026 race has three fronts: capability, cost, and who audits the model before it reaches the public.