OpenAI's three-tier frontier model family — Sol, Terra, and Luna — with sharper image reading and stronger text-and-interface generation.
Last verified · 2026-07-11 · by Moe Ameen
GPT-5.6 is OpenAI's frontier model generation, released as a family of three rather than a single model. Sol is the flagship for the hardest reasoning and coding; Terra is the balanced everyday tier, roughly GPT-5.5-class quality at about half the price; and Luna is the fastest and cheapest. OpenAI previewed the family to a small group of partners around June 26, 2026 and made it generally available across the OpenAI API, ChatGPT, and Codex on July 9, 2026. API pricing runs $5 per million input tokens and $30 per million output for Sol, $2.50/$15 for Terra, and $1/$6 for Luna.
What OpenAI improved most for content people is multimodal reading. All three tiers accept text and image input and return text, and GPT-5.6 handles reference images more faithfully — a new "detail: original" image setting preserves what you paste in, so the model can reason over a screenshot, a competitor's post, a product photo, or a page layout instead of paraphrasing it. On top of that, GPT-5.6 is markedly better at generating "artifacts" — cohesive, ready-to-use outputs like structured briefs, tables, and even tasteful interface mockups from only a high-level direction. It carries a roughly million-token context window, up to 128,000 output tokens, and a February 16, 2026 knowledge cutoff, plus new API features: Programmatic Tool Calling (it writes JavaScript that runs in an isolated sandbox) and a subagent-powered "ultra" mode on Sol.
The honest boundary matters, because "improved multimodal" gets read as "it makes images and video now." It does not. GPT-5.6 reads images and outputs text (and code) — it is a reasoning-and-writing brain, not a media generator. It renders no video, draws no images, designs no branded graphics, holds no persistent brand system, and publishes to nothing. If you found it looking for a way to make and post content, it is the upstream layer that thinks and drafts; the making and shipping is a separate job. (For its agentic coding surface, see GPT-5.6 Sol Ultra in Codex.)
The useful way to think about GPT-5.6 for content is as a reader-and-writer that stops one step short of anything publishable. Paste a product photo or a rival's carousel into GPT-5.6 and it will genuinely reason over the image and hand you a sharp script, a set of hooks, or a repurposing plan. What it cannot do is turn that plan into the actual pixels — the talking-head video, the branded carousel, the quote card — or put any of it on your channels. That gap is precisely where Kompozy takes over. Feed the script or angle GPT-5.6 produced into Kompozy as a source and it renders the media the model could only describe: a Persona Short where your face-locked avatar delivers the script with burned-in captions, brand-exact Carousel Posts and Quote Graphics through HyperFrames, Photo Posts and Infographics, plus a Blog Article and an Email Newsletter — every piece held to one voice by your Persona Brief.
So the handoff is clean: GPT-5.6 reads the reference and thinks; Kompozy generates the finished formats and fans them across nine social platforms plus email and blog from a single queue, on Autopilot with a per-post review pipeline. Worth knowing — Kompozy runs its own copy generation on managed OpenAI and Claude models, this same class of frontier model, so you get GPT-5.6-grade writing inside the engine without wiring an API, picking a tier, or paying per token yourself. The model reads and drafts; the engine renders and ships.
GPT-5.6 is OpenAI's frontier model family, released as three tiers: Sol (flagship), Terra (balanced, about half the price of GPT-5.5), and Luna (fast and cheapest). It went generally available across the OpenAI API, ChatGPT, and Codex on July 9, 2026 after a late-June preview. It accepts text and image input and returns text.
No. Despite "improved multimodal," GPT-5.6 reads images and outputs text and code — it does not generate images, video, or audio. To turn what it drafts into finished, published media, pair it with a content engine like Kompozy that renders persona video, carousels, images, blogs, and newsletters and publishes them across platforms.
They are the three tiers of GPT-5.6. Sol is the flagship for the hardest work ($5/$30 per million input/output tokens); Terra is the balanced everyday tier at roughly GPT-5.5-class quality for about half the price ($2.50/$15); Luna is the fastest and cheapest ($1/$6). All three share the same multimodal reading and roughly million-token context window.
On the API, Sol is $5 per million input tokens and $30 per million output; Terra is $2.50/$15 and Luna is $1/$6. In ChatGPT, GPT-5.6 is available on the paid plans (which can select Sol); the free tier defaults to the older GPT-5.5 Instant. Kompozy, by contrast, runs generation on managed models inside a flat subscription, so you do not pay per token.
You can draft on it, but not finish or publish on it. GPT-5.6 produces text; a content workflow also needs media rendering, a brand system, design, scheduling, and multi-platform publishing — none of which a raw model does. Kompozy is that surrounding engine, and it already generates on this class of model under the hood.