OpenAI has just shipped GPT-5.6 — but you probably can't log in and use it yet. The release is a limited preview available to a small set of trusted partners, delivered through the API and Codex rather than the ChatGPT app. What is public, though, is the full lineup and the official pricing: three named tiers — Sol, Terra, and Luna — each with its own price, context window, and target workload.
This guide uses OpenAI's own developer docs (linked at the end) to break down what each tier is for, what it costs versus GPT-5.5, how it differs, and the realistic ways to get access before broad rollout.
What is GPT-5.6?
GPT-5.6 is OpenAI's newest flagship series, and it drops the old mini / nano suffixes for three named tiers that signal intelligence, speed, and cost:
Sol — the flagship, tuned for complex coding, long-horizon agentic tasks, and safety research.
Terra — the balanced tier, positioned to match GPT-5.5's quality at roughly half the price.
Luna — the low-cost tier, built for fast, high-volume work like summarizing, drafting, and tagging.
OpenAI's preview announcement is titled "Previewing GPT-5.6 Sol," which tells you where the emphasis sits. Per OpenAI's developer docs, the latest models take text and image input, return text output, and support vision, served through the Responses API and the official SDKs. Note that image input means GPT-5.6 can read images — OpenAI has not announced native image generation for this series.
Sol, Terra, Luna: which tier to choose
The three tiers map to different jobs. OpenAI's docs also list two reasoning modes on top: max (more reasoning effort) and ultra (sub-agent collaboration for more complex multi-step workflows).
Here's the spec sheet, taken from OpenAI's developer model pages:
Tier | Model ID | Input ($/MTok) | Output ($/MTok) | Context | Max output | Knowledge cutoff |
|---|---|---|---|---|---|---|
Sol |
| $5.00 | $30.00 | 1M | 128K | Dec 1, 2025 |
Terra |
| $2.50 | $15.00 | 1M | 128K | Aug 31, 2025 |
Luna |
| $0.75 | $4.50 | 400K | 128K | — |
Rules of thumb:
Pick Sol for hard engineering work — refactors, multi-step automation, agentic pipelines, or security research where accuracy matters more than cost.
Pick Terra when you want GPT-5.5-class quality for customer support, internal tools, or document analysis but need to control spend at scale.
Pick Luna for high-throughput, latency-sensitive jobs — batch first drafts, content tagging, summaries — where "good and fast" beats "best."
GPT-5.6 vs GPT-5.5: price and positioning
The clearest way to read GPT-5.6 is against GPT-5.5, which lists at $5 / $30 per million tokens (input / output) in the same docs:
GPT-5.5 | GPT-5.6 Terra | GPT-5.6 Sol | |
|---|---|---|---|
Input / Output ($/MTok) | $5 / $30 | $2.50 / $15 | $5 / $30 |
vs GPT-5.5 price | baseline | exactly half | same |
Context | 1M | 1M | 1M |
Positioning | Prior flagship | ≈5.5 quality | New flagship |
Best for | General reasoning/coding | Cost-sensitive scale | Hard coding, agentic |
GPT-5.6 doesn't raise the ceiling price — Sol matches 5.5 — it adds cheaper tiers underneath. Terra is the headline: half of GPT-5.5's rate, which OpenAI positions at 5.5-class quality (its own claim, not an independent benchmark). Sol sits above 5.5 on harder work — OpenAI's preview notes cite gains on agentic and command-line coding and more token-efficient long-horizon tasks, but it hasn't published raw scores yet, so treat those as stated-not-quantified for now. So if you're on 5.5: Terra is the cost-down move, Sol is the capability-up move.
A concrete cost example — a task that sends 1M input tokens and returns 200K output tokens:
Tier | Input | Output | Total |
|---|---|---|---|
Sol | $5.00 | $6.00 | $11.00 |
Terra | $2.50 | $3.00 | $5.50 |
Luna | $0.75 | $0.90 | $1.65 |
The series also supports prompt caching, so reused context (cached input) is billed at a discount to the standard input rate. Two corrections worth flagging, because they spread through early coverage: several write-ups listed Luna at $1.00 / $6.00 — the documented rate is $0.75 / $4.50 — and the repeated "1.5M token context" doesn't match the docs, which show 1M for Sol/Terra and 400K for Luna.
Can you use GPT-5.6 yet?
Short answer: not unless you're a preview partner. Searches like "gpt 5.6 banned" are mostly a misreading — GPT-5.6 isn't banned, it's gated. Access runs through the API and Codex only, the models are not yet in ChatGPT, and OpenAI's docs say broad availability is "coming soon" without a public date.
The series also ships with layered safety controls — model-level refusals, real-time detection, tiered access, and ongoing red-teaming. Combined with the partner-only gate, that's what's fueling the "is it restricted?" chatter.
How to access GPT-5.6
Start with the official check: list the models available to your account and see whether the gpt-5.6-* IDs come back.
curl https://api.openai.com/v1/models \
-H "Authorization: Bearer $OPENAI_API_KEY" | grep gpt-5.6If gpt-5.6-sol / -terra / -luna appear, you're in the preview and can call them via the Responses API or Codex. If you get an empty result — or a model_not_found / permission error when you call one — your account isn't in the preview yet, and the honest answer is to wait for the broad rollout OpenAI says is coming. In the meantime, GPT-5.5 is the closest drop-in fallback for anything you'd have routed to 5.6 Terra.
FAQ
When will GPT-5.6 be available to everyone?
OpenAI only says broad API availability is "coming soon" and hasn't given a public date. ChatGPT access typically follows API access, so expect it there later than for developers.
Which tier should I default to?
Terra for most production traffic (5.5-class quality at half the price), Sol when accuracy on hard coding/agentic tasks outweighs cost, Luna for cheap high-volume drafting and tagging.
Does GPT-5.6 have a free tier?
No. It's a closed, paid model in preview; there's no free API allowance, and it isn't in the free ChatGPT tiers.
How is GPT-5.6 different from GPT-5.5?
Terra matches 5.5's quality for half the price, while Sol pushes ahead on coding and long-horizon agentic tasks. New max and ultra reasoning modes also debut here.
Can GPT-5.6 generate images?
It accepts image input (vision) but OpenAI has not announced native image generation for the 5.6 series.
Sources
OpenAI — Models (developer docs): https://developers.openai.com/api/docs/models
OpenAI — GPT-5.6 Sol model page (pricing, context, cutoff): https://developers.openai.com/api/docs/models/gpt-5.6-sol
OpenAI — GPT-5.6 Terra model page: https://developers.openai.com/api/docs/models/gpt-5.6-terra
OpenAI — GPT-5.6 Luna model page: https://developers.openai.com/api/docs/models/gpt-5.6-luna
OpenAI — GPT-5.5 model page (price baseline): https://developers.openai.com/api/docs/models/gpt-5.5
OpenAI — "Previewing GPT-5.6 Sol" announcement: https://openai.com/index/previewing-gpt-5-6-sol/
Pricing and specs verified against OpenAI's developer docs at the preview launch. Raw benchmark scores and the broad-availability date will be added once OpenAI publishes them.