OpenAI's cost-optimized GPT-6 models — Sol for complex coding and knowledge work, Luna for cheap high-volume tasks — launched September 22, 2026 with fewer mistakes at roughly half the price of the GPT-5.6 tier.
Last verified · 2026-09-22 · by Moe Ameen
GPT-6 Sol and GPT-6 Luna are the cost-optimized models in OpenAI's GPT-6 family, launched September 22, 2026 after the flagship GPT-6 Astra earlier in the month. OpenAI positioned them around a plain promise rather than a new capability: lower cost and fewer mistakes. Both ship at roughly half the price of their GPT-5.6 namesakes — OpenAI listed Sol at $2 per million input tokens and $10 per million output tokens (down from $4 and $20), and Luna at $0.10 and $0.50 (down from $0.20 and $1.20) — and described the prices as permanent, not promotional, crediting caching and inference improvements.
The two models are built for different jobs. Sol is aimed at complex, repeated knowledge work — building features, reviewing code, debugging, and analyzing data. Luna is the high-volume option for tightly-scoped tasks like summarizing documents, extracting information, and answering quick questions. On OpenAI's internal factuality evaluation, which it describes as drawn from de-identified real conversations where users flagged model mistakes, OpenAI said Sol makes roughly half as many mistakes as GPT-5.6 Sol — approaching flagship Astra reliability at much lower cost — and reported a lower coding error rate.
Both 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. One thing to be clear about: Sol and Luna are text and reasoning models. They write, reason, and code — they generate no images, no video, no captions, and they publish to no platform. Treat exact pricing, benchmark figures, and rollout details as a launch-window snapshot and confirm them against OpenAI's own pages.
Sol and Luna are built for two different jobs, and that split maps cleanly onto how [Kompozy](/) uses a model. Luna is the cheap, high-volume workhorse — ideal for generating the bulk copy a content week needs: dozens of caption variations, per-platform rewrites, a source summarized into ten post angles. Sol is the one you reach for when the draft has to reason — a tight [Persona Short](/glossary/persona-shorts) script, a structured blog outline, a synthesis of a rambling transcript. Neither, though, turns any of that into something you can post. That is the handoff Kompozy is built for.
Concretely: draft in Sol or Luna, then bring the text into Kompozy as your source. Kompozy rewrites it through a [Persona Brief](/glossary/persona-brief) so it reads like you, then generates the finished formats a model can't — a captioned Persona Short or a HeyGen avatar video narrating the script, a brand-exact [Carousel](/glossary/hyperframes) built from the outline, Quote Graphics pulling the sharpest lines, [Clipped Shorts](/glossary/clipped-short) from a longer cut, per-platform Text Posts, a full Blog Article, and an Email Newsletter across [18 formats](/glossary/output-buckets). Then [Autopilot](/glossary/autopilot) schedules and publishes the set across the eight social platforms plus blog and email behind a per-post review. Because Luna is so cheap you can afford to over-generate raw copy and let Kompozy shape the best of it; because Sol is more accurate, the scripts it hands off need less fixing. The model thinks and drafts; Kompozy makes and ships.
They are the 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 cheaper, faster model for high-volume, tightly-scoped tasks like summarization and extraction. OpenAI positions both around lower cost and fewer mistakes.
Sol is the higher-capability model for complex, repeated work — building features, reviewing code, debugging, and analysis. Luna is a cheaper, high-volume model for straightforward jobs like summarizing, extracting, and answering quick questions. Sol costs more per token; Luna is priced to run at scale.
No. Both are text and reasoning models — they draft, code, and analyze but produce no video, images, or audio, and publish to no platform. To turn their drafts into avatar video, carousels, quote graphics, and other finished media, pair them with a content engine like Kompozy that generates those formats and publishes them.
Use Luna to generate bulk copy or Sol to draft the reasoning-heavy piece, then bring that into Kompozy. Kompozy rewrites it in your voice via a Persona Brief, generates Shorts, carousels, quote graphics, text posts, a blog, and a newsletter, and schedules and publishes them across the eight social platforms plus blog and email.
OpenAI listed Sol at $2 per million input tokens and $10 per million output tokens, and Luna at $0.10 and $0.50 — about half their GPT-5.6 counterparts, and described as permanent rather than promotional prices. Confirm current figures on OpenAI's own pages, as pricing and availability vary by plan and region.