Announced September 22, 2026, GPT-6 Sol and GPT-6 Luna are cost-optimized additions to the GPT-6 family — priced at roughly half their GPT-5.6 counterparts, with OpenAI claiming Sol makes about half as many mistakes as its predecessor.
2026-09-22 · by Moe Ameen
On September 22, 2026, OpenAI introduced GPT-6 Sol and GPT-6 Luna, two cost-optimized models in the GPT-6 family that began earlier in the month with the flagship GPT-6 Astra. The pitch is deliberately unflashy: fewer mistakes at lower cost, rather than a new headline capability. Both ship at roughly half the price of their GPT-5.6 namesakes. OpenAI listed GPT-6 Sol at $2 per million input tokens and $10 per million output tokens (down from $4 and $20 on GPT-5.6 Sol), and GPT-6 Luna at $0.10 per million input tokens and $0.50 per million output tokens (down from $0.20 and $1.20). The company said these are permanent prices — not promotional or introductory — and credited the drop to caching and inference improvements.
The two models split by job. Sol is aimed at complex, repeated knowledge work — building features, reviewing code, debugging, and analyzing data. Luna is the high-volume option for more tightly defined tasks like summarizing documents, extracting information, and answering straightforward questions. On OpenAI's internal factuality evaluation, which it describes as drawn from de-identified real-world conversations where users flagged model mistakes, OpenAI said GPT-6 Sol makes roughly half as many mistakes as GPT-5.6 Sol — approaching the reliability of the flagship Astra at much lower cost — and reported a lower error rate on coding as well.
Both models are available through the OpenAI API and are rolling out in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, with Luna also reaching the desktop app and free tiers. OpenAI shared coding and agentic-task benchmark figures alongside the launch. Treat the exact benchmark numbers, pricing, and rollout timing as a launch-window snapshot and confirm them against OpenAI's own pages before relying on them.
The news for creators is not a new capability — it is cheaper, more accurate text. That lands at the copy layer of a content pipeline, and it is exactly where [Kompozy](/) already puts an OpenAI or Claude model to work. Kompozy uses these models to draft the words — captions, text posts, blog articles, newsletter bodies — governed by a [Persona Brief](/glossary/persona-brief) so the output reads like you and respects your banned-word list. A model that makes half as many mistakes means fewer weak drafts reaching Kompozy's per-post review gate, and on the Founding tier, where you bring your own key, Luna's lower price directly cuts the cost of generating text in bulk. The practical move today is to let the cheaper, sharper model handle the drafting and keep your attention on approving output.
But a cheaper, sharper model does not ship a post — it returns text. Kompozy is the layer that turns that text into finished, published content: from one idea it generates [Persona Shorts](/glossary/persona-shorts), [Clipped Shorts](/glossary/clipped-short), brand-exact [Carousel Posts](/glossary/hyperframes), Quote Graphics, and full blogs and newsletters across [18 formats](/glossary/output-buckets), then schedules and publishes them across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot). When the model underneath gets cheaper and more reliable, the economics of the whole pipeline improve — but the thing that decides whether you actually post consistently is still the engine that produces and distributes the content, not the tokens beneath it.
They are two cost-optimized models in OpenAI's GPT-6 family, launched September 22, 2026 alongside the earlier flagship GPT-6 Astra. Sol is aimed at complex coding and knowledge work; Luna is a high-volume model for tightly-scoped tasks like summarization, extraction, and quick questions. OpenAI positions both around lower cost and fewer mistakes rather than a new headline capability.
OpenAI listed GPT-6 Sol at $2 per million input tokens and $10 per million output tokens, and GPT-6 Luna at $0.10 and $0.50 — roughly half the price of their GPT-5.6 counterparts, and described as permanent prices rather than promotional. OpenAI credited caching and inference improvements. Confirm current figures on OpenAI's own pricing pages.
OpenAI says GPT-6 Sol makes roughly half as many mistakes as GPT-5.6 Sol on its internal factuality evaluation — which it describes as based on de-identified real conversations where users flagged model errors — and reports a lower coding error rate. It positions Sol as approaching the reliability of the flagship GPT-6 Astra at much lower cost.
No. Sol and Luna are text and reasoning models — they draft, code, and analyze, but generate no video or images and publish to no platform. To turn their drafts into finished, scheduled posts across platforms, you pair them with a content engine like Kompozy that generates video, carousels, and posts from one idea and publishes across nine platforms.