GPT-5.6 Is Now Generally Available: Fewer Tokens, Lower Cost
OpenAI has made its GPT-5.6 family — which it opened only to a narrow group last month — generally available. There are three models: flagship Sol, the balanced Terra for everyday work, and the most affordable Luna. The announcement is less about raw performance than efficiency: OpenAI says GPT-5.6 does the same work with fewer tokens, in less time, and at lower cost, beating rival flagships on many tests.
In brief
- What happened: GPT-5.6 (Sol/Terra/Luna) moved out of preview into general availability. It's in ChatGPT, Codex, and the API from today, rolling out worldwide over 24 hours. Free and Go users get Terra.
- The efficiency claim: per OpenAI, Sol tops the coding-agent index at 80 and beats Claude Fable 5 on many tests using fewer tokens at about a third of the cost. But it isn't first everywhere — Fable 5 and Mythos lead on SWE-Bench Pro and advanced math, for instance. The scores are OpenAI's; some are drawn from independent indexes (Artificial Analysis).
- What it's good at: coding, office work (slides, docs, spreadsheets), UI design, and computer use stand out. An "ultra" setting runs four agents in parallel on hard tasks.
- Price + safety: per 1M tokens, Sol $5/$30, Terra $2.50/$15, Luna $1/$6. OpenAI says GPT-5.6 does not cross the "critical threshold" in cyber or biology and ships with its strongest safeguards yet.
Our take
The real message for you is, again, the bill: GPT-5.6's pitch isn't to be the smartest model so much as to do the same work with fewer tokens and for less. If that holds up, it shows up directly in your monthly AI cost. And it's not just for coders — it produces tidier results in office work like slides, docs, and spreadsheets, which is most people's day job. Free and Go users can reach it via Terra, so trying it doesn't require an upgrade. Two notes: most of the scores are OpenAI's own numbers, so don't take them on faith — and the model isn't first on every test, so depending on your task another model may be better. Still, the "same work for less" angle is a bar anyone trying to cut costs should look at.
Kaynak: OpenAI