OpenAI shipped GPT-5.6 on July 9, 2026, and this time it is not one model but three: Sol, Terra, and Luna. All three are already wired into Playcode's AI website builder. Here is a plain-English guide to what each one does, what it costs, and which to pick for the page in front of you.
One generation, three tiers
OpenAI changed how it names models. The number (5.6) is the generation; Sol, Terra, and Luna are durable capability tiers that can each improve on their own schedule. In OpenAI's own words: "The number identifies the generation, while Sol, Terra, and Luna are durable capability tiers that can advance on their own cadence."
- GPT-5.6 Sol - the flagship. Highest quality of the family, best design judgment, strongest on the hardest builds.
- GPT-5.6 Terra - the balanced everyday model. OpenAI positions it as roughly GPT-5.5-level quality at about half the cost. This is the new default GPT for most website work.
- GPT-5.6 Luna - the fast, cheap tier. Built for high-volume, latency- sensitive work: quick edits, first-pass drafts, simple pages.
All three share the same big specs: a ~1 million token context window, up to 128K tokens of output, and a February 2026 knowledge cutoff. The differences are quality, speed, and price.
What each one costs
OpenAI's published rate card, per 1M tokens:
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-5.6 Sol | $5.00 | $0.50 | $30.00 |
| GPT-5.6 Terra | $2.50 | $0.25 | $15.00 |
| GPT-5.6 Luna | $1.00 | $0.10 | $6.00 |
In Playcode you never do this math yourself. Each model shows a credit-cost badge in the picker, so you can see the trade before you pick. The practical read: Terra costs about half of Sol per request, and Luna is cheaper still. Two honest footnotes on the pricing - from GPT-5.6 on, a cache write costs 1.25x the normal input rate (cache reads still get the usual 90% discount), and prompts above ~272K tokens bill the whole request at double the input rate. Neither matters for a typical landing page; both matter if you point the model at an entire large codebase.
What is actually new (per OpenAI)
- Efficiency is the headline. GPT-5.6 reaches frontier results with far fewer tokens. Sam Altman put the agentic-coding token savings at around 54% versus the previous generation. Fewer tokens means lower cost and lower latency for the same task.
- A real step up in design judgment. OpenAI is explicit that this generation was tuned for front-end taste: "With only high-level direction, GPT-5.6 creates tasteful, ergonomic, and functional interfaces." It can also inspect the rendered result and refine it, not just emit code and hope.
- A million-token context window. Big enough to hold a substantial project in view, so the model stops losing track of files across a long build.
- Better tool use. More precise tool selection on large tool surfaces, which matters for an agent like Playcode's that has many tools to choose from.
Which tier for building websites?
Our recommendation, and how we have set the picker up:
- Start on Terra. It clears the vast majority of real website work - landing pages, marketing sites, dashboards - at a cost that fits the credit economy on a Pro plan. It is the sensible default GPT.
- Reach for Sol when a build is genuinely hard or when design quality is the whole point: a hero section for a paying client, a page you will show your team. Sol has the best taste and reasoning of the three.
- Drop to Luna for quick, cheap passes. Small copy tweaks, a fast first draft, a simple page you will refine by hand. You would rather iterate five times cheaply than wait once expensively.
An honest take: efficiency, not a clean sweep
GPT-5.6 is a strong release, but do not read the launch as "OpenAI now wins everything." On OpenAI's own published benchmarks, Anthropic's Claude models still lead on SWE-Bench Pro (a hard real-world coding benchmark) by a wide margin, and on a couple of the raw-intelligence indexes. Where GPT-5.6 clearly shines is the combination of strong results and much lower token cost, plus genuinely improved design sense. Simon Willison, who had early access, summed up the measured view: it is "definitely very competent, though so far it hasn't struck me as better than Fable at the kind of complex coding tasks" he runs on Anthropic's model.
That is exactly why Playcode keeps every strong model one click apart instead of crowning a single winner. The right model depends on what is on your screen right now.
How to select GPT-5.6 in Playcode
Open any Playcode project, then:
- Click the Menu
- Open Preferences
- Pick AI Model
- Choose GPT-5.6 Terra, Sol, or Luna
The best habit with any new model: run the same prompt on your usual pick and on the new one, then compare the actual result. You will usually have a clear preference within thirty seconds. Switching back takes ten.
Open Playcode at playcode.io and pick your model under Preferences. New users get free credits to try the AI website builder before subscribing.