July 11, 20267 min read

GPT-5.6 Is Here: Sol, Terra, Luna — Efficiency by Default, Power on Demand

OpenAI launched GPT-5.6 on July 9, 2026 across ChatGPT, Codex, and the API: flagship Sol, balanced Terra, and cost-efficient Luna. Here is what shipped, the new pricing tiers, and what it means if you build with AI.

GPT-5.6OpenAIAI models

On July 9, 2026, OpenAI launched the GPT-5.6 family for general availability across ChatGPT, Codex, and the OpenAI API. After a limited preview that began on June 26, the full rollout is now underway: Sol as the new flagship, Terra as a balanced everyday tier, and Luna as the cheapest option in the line.

After Anthropic shipped Fable, the U.S. government tightened terms around global access to AI services, so OpenAI had to open GPT-5.6 to a small set of users first and only go global once that looked safe. People were fairly excited about how it would land against Fable’s punch — so far the community read is that it doesn’t really land that punch: mostly more expensive, without a clear efficiency jump over 5.5.

What OpenAI shipped

According to OpenAI's official GPT-5.6 announcement, ChatGPT, Codex, and the API went live together on July 9, with global availability filling in over roughly the next 24 hours.

The naming is intentional:

Tier	Role	Official list price (per 1M tokens)
Sol	Flagship intelligence and coding	$5 input / $30 output
Terra	Balanced everyday work; competitive with GPT-5.5 at lower cost — roughly the old “mini” slot	$2.50 input / $15 output
Luna	Fastest and cheapest in the family — roughly the old “nano” slot	$1 input / $6 output

The number (5.6) marks the generation. Sol, Terra, and Luna are durable capability tiers that can advance on their own cadence. In the API, the explicit model IDs are gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna. The short name gpt-5.6 is equivalent to Sol — use the full ID if you want Terra or Luna.

Why this release is different

OpenAI frames GPT-5.6 as efficient by default, maximum performance on demand. Below is the official story:

Area	What OpenAI highlighted	Why it matters
Token efficiency	Stronger results with fewer tokens and lower estimated cost vs prior and competing models	Cost per finished task can look better than list price alone
Coding agents	SOTA claims on Artificial Analysis Coding Agent Index, Terminal-Bench 2.1, DeepSWE	Better fit for multi-step implementation, CLI workflows, and long-horizon engineering
Knowledge work	Stronger BrowseComp and OSWorld 2.0; more solid decks, docs, and spreadsheets	More reliable path from messy inputs to deliverables
Design + computer use	Inspect and refine rendered UI, not only emit code	Closer to a collaborator that can hand work back
Programmatic Tool Calling	Model can write and run light programs that coordinate tools	Fewer round trips and less intermediate token bloat
`max` / `ultra`	Longer reasoning time (`max`); multi-agent parallel work by default (`ultra`)	Spend more only when the task is worth it

Partners piled on with the same line: better quality, often fewer tokens, lower latency. Treat that commercial mutual praise as optional reading — try it yourself.

Availability and API details

In this product line, GPT-5.6 is the first time ChatGPT, Codex, and the API shipped on the same day (after only a short trusted-partner preview).

ChatGPT / ChatGPT Work / Codex: Sol / Terra / Luna roll out by plan; Free and Go usually see a more limited set than Plus / Pro / Business / Enterprise. Some people in community chats report they still don’t see 5.6 at all.
API: All three tiers are callable. Programmatic Tool Calling in the Responses API is marked Zero Data Retention (ZDR) compatible; multi-agent starts in beta for ultra-like coordination.
Prompt caching: For GPT-5.6 and later, cache writes bill at 1.25× the rate; cache reads still get a 90% off (one-tenth price) discount. Using prompt caching needs explicit cache breakpoints and at least a 30-minute cache lifetime.

Before production, trust your account’s model picker and the API pricing page — don’t hard-code third-party context lengths. Many third-party lists say Sol/Terra are about 1M context; treat that as a hint until your own dashboard agrees.

Safety, awkwardly

OpenAI says GPT-5.6 ships with its strongest safeguards yet, after a longer evaluation cycle: red teaming, large-scale automated tests, and partner pressure-testing during the preview. The more you expect the model to do work, the more control matters — especially when you wire it into agents that can browse, edit code, or operate software.

Unfortunately, people were already complaining on X today that Codex deleted their entire home directory…

Why this matters for builders

GPT-5.6 keeps pushing the industry from “smart chat” toward operators that finish work. The practical questions become:

Which tier for which job: Sol for hard agentic work, Terra as the daily default, Luna for volume and cost?
Do you evaluate cost as “price of a finished workflow,” not only per-token list price?
When is ultra worth it vs a single-agent max pass?
How do you keep integration stable while model names and tiers keep moving?

That last point is exactly where a unified gateway earns its keep.

What this means for MuiRouter users

MuiRouter’s goal is simple: one API key, one request shape, and access to the best current models. OpenAI has shipped officially, so you can start trying these new models as well.

Bottom line

GPT-5.6 is worth watching because OpenAI is explicit about cost to finish a task, not cost per token. If you use ChatGPT or Codex, try Sol first on long agentic and coding work; for everyday copy and light work, Terra is the sensible first tier. Luna — I personally wouldn’t recommend leaning on the lowest tier.

From early community feedback, the jump over 5.5 may not be dramatic, and it doesn’t clear Fable either. Treat it as a routine knowledge-base upgrade for now.

Reference sources

Primary source published on July 9, 2026.

Be ready for the next AI shift

Start with one API key and a cleaner path to keep model access stable as tools and upstream availability change.