GPT-5.6 Sol, Luna and Terra: OpenAI's August reshuffle, explained
In late July and early August 2026, OpenAI reshaped the GPT-5.6 family on both price and capability: GPT-5.6 Luna API prices dropped 80% (Terra 20%), the updated GPT-5.6 Sol cut factual errors by 68% in finance, medicine and law evaluations, and free-tier users gained unlimited Luna text chats plus the extended-thinking "Think" button.
What are Sol, Luna and Terra?
GPT-5.6 ships as three variants. Based on OpenAI's announcements tracked on our GPT-5.6 model page: Sol is the flagship — it now powers all chats for paid ChatGPT users, handling both instant responses and deep reasoning in one consistent experience. Luna is the efficient tier — now the free-tier default with unlimited text chats. Terra appears primarily as an API-tier model, referenced in OpenAI's pricing announcements.
The changes, dated
| Date | Change | Who it affects |
|---|---|---|
| Jul 30 | Luna API price −80%, Terra −20%, faster Sol option in API | API developers |
| Aug 6 | Updated Sol powers all chats (Instant + deep reasoning); 68% fewer factual-error responses than GPT-5.5 Instant in high-stakes eval | Plus & Pro |
| Aug 6 | Reasoning-effort slider for everyday chats | Plus & Pro |
| Aug 6 | Luna upgrade + "Think" extended reasoning button | Free & Go |
| Aug 7 | Unlimited Luna text chats begin rolling out | Free & Go |
Why does the 68% factuality number matter?
OpenAI's evaluation covered finance, medicine and law — the domains where a wrong answer costs money or safety, and where enterprises have been slowest to deploy chatbots. The comparison baseline is GPT-5.5 Instant, the previous everyday-chat default. Cutting factual errors by two-thirds in those domains is OpenAI's case that the everyday model is now trustworthy enough for high-stakes use without routing to a slower reasoning mode — which is also why Sol now handles Instant and deep reasoning as a single model rather than two experiences.
What's the strategic read?
Price and access moved in one direction: down and out. An 80% API price cut on Luna, unlimited free-tier chats, and extended thinking for non-paying users all landed within eight days — the same window in which Alibaba shipped Qwen 3.8-Max "better and cheaper" with open weights promised, and Tencent's Hy3 topped the OpenRouter leaderboard. The frontier-model market is now a price war at the low end and a factuality war at the high end, and OpenAI is fighting both at once.