The short answer
OpenAI released GPT-5.6 publicly on 9 July 2026 in three tiers — Sol (flagship), Terra (everyday) and Luna (lowest cost) — after a roughly two-week preview limited to vetted partners while the US government reviewed the models. Reported API rates are $5/$30 per million tokens for Sol, $2.50/$15 for Terra and $1/$6 for Luna, with Terra pitched as near the prior generation at about half the cost.
For teams building software for US and EU markets, there are two durable signals. The tier spread — roughly 5× between top and bottom — makes which tier runs each step an explicit routing decision, not a default. And the pre-release review sets a precedent: frontier availability can now be gated or delayed by a process you do not control, so build for release timing you cannot assume.
What happened on 9 July?
OpenAI made GPT-5.6 generally available on Thursday, 9 July 2026, publishing three variants across ChatGPT and the API: Sol, the strongest of the family; Terra, positioned for everyday use at roughly the prior generation's quality for about half the cost; and Luna, the lowest-cost option. The release ended a gated period that began in late June, when the models were shared only with a small group of vetted partner organisations — a departure from OpenAI's usual ship-then-iterate cadence — before access was, in the company's words, expanded globally.
The concrete numbers teams will plan around are the API rates. Per multiple reports at launch, Sol lists at $5 per million input tokens and $30 per million output; Terra at $2.50 and $15; and Luna at $1 and $6. That is a spread of roughly 5× from the top tier to the bottom for the same token — and it lands in a market where, since OpenAI and its rivals resumed shipping, every major frontier lab again has a publicly available model at once. Abundance and a wide price ladder are the backdrop; how you pick among the rungs is the decision.
Treat the exact figures as a moving target — frontier rates and rankings shift within weeks — and the structure as the durable part: a premium flagship, a strong mid-tier at a fraction of the cost, and a cheap tier for the high-volume, low-difficulty work that dominates most real workloads.
The three tiers, in one table
Here is the shape of the choice GPT-5.6 puts in front of teams. Rates are per million tokens and move often; confirm with the provider before you commit a budget.
| Tier | Reported rate (in / out) | Best-fit work |
|---|---|---|
| Sol (flagship) | $5 / $30 | Hardest planning, ambiguous reasoning, complex code |
| Terra (everyday) | $2.50 / $15 | General reasoning, drafting, tool orchestration at volume |
| Luna (lowest cost) | $1 / $6 | Classification, extraction, tagging, formatting, summarisation |
The point of the table is not the exact numbers, which will be stale within weeks. It is the ratio: an autonomous loop that plans, calls tools, retries and formats can burn millions of tokens, and pointing all of that at the flagship scales cost with the loop rather than the value delivered. A run that costs a few dollars on Luna can cost many times that on Sol for output that a cheaper tier would have handled just as well.
Why did a government review it first?
The gated rollout was not a supply problem; it was policy. An AI cybersecurity order signed in early June 2026 asked companies to voluntarily present their most powerful models to the US government roughly 30 days before public release. OpenAI ran the limited preview with vetted partners while the Department of Commerce's Center for AI Standards and Innovation carried out additional testing, and sent technical staff to Washington for meetings, before receiving permission for a wider launch. The company was candid that it complied to move faster rather than out of enthusiasm, stating it does not believe “this kind of government access process should become the long-term default.”
Whatever one makes of the policy, the operational takeaway is neutral and concrete: the moment a frontier model becomes available to build on is now something an external process can gate or delay. For a US team that intersects with the EU AI Act on the other side of the Atlantic, the broader pattern is the same — model availability and model governance are converging, and both belong in your planning rather than your surprises.
What it means for US & EU software teams
Strip away the launch-day noise and two design signals remain. First, build tier routing in, behind an abstraction. Put a provider-agnostic layer between your business logic and any model so switching or mixing tiers — and providers — is configuration, not a rewrite; route hard reasoning, planning and code to Sol, everyday work to Terra, and high-volume simple steps to Luna or an open-weight model; and instrument cost and quality per step so the routing is tuned by evidence. That orchestration layer is exactly what GenAI integration work is for, and it is what turns a new tier from a migration into a config edit.
Second, plan for availability you do not control. A model you intend to depend on may arrive on a schedule set partly by a review process, and a critical model can be gated, delayed or region-limited. Keep at least one alternate model wired up so your roadmap does not hinge on a single launch date, and record which model version you shipped and what testing it passed — increasingly part of vendor-risk and audit trails. In regulated verticals such as FinTech, “we depend on one model, released on someone else's timeline, with no fallback” is a concentration risk an auditor will name; a routing layer with a configured backup is both cheaper and more resilient.
None of this means chasing the cheapest tier everywhere. Terra or an open-weight model that lands within a few points of the flagship on routine steps is the right call for those steps; Sol earns its rate on the genuinely hard ones. The discipline is matching capability to difficulty — and keeping the freedom to change your mind cheaply as the next release, or the next review, lands.
A practical readiness checklist
Nothing here is a new deadline. It is the work that turns a model launch — gated or not — into a shrug:
- Add an abstraction layer. Route every model call through one internal interface so provider, model and tier are configuration, not code scattered across the app.
- Classify your steps. Split work into hard (planning, ambiguous reasoning, complex code) and routine (classification, extraction, drafting, formatting), and assign a tier to each.
- Instrument per-step cost and quality. Log tokens, latency and an eval score per step so you can see where money and errors actually go before you scale.
- Configure a fallback model. Keep at least one alternate provider wired up so a gated launch, price change, quota limit or outage degrades gracefully.
- Record model provenance. Note which model version and what testing status you shipped, so vendor-risk and compliance reviews are a lookup, not a scramble.
- Re-evaluate on a cadence. Tiers, prices and rankings move monthly; revisit routing on a schedule rather than treating the first choice as permanent.
This is not investment or procurement advice, and the right tier mix depends on your workloads, your quality bar and your markets. But the signal from GPT-5.6's launch is clear: the frontier now ships in graded tiers, on timelines shaped by review, so build the machinery to spend the top tier only where it earns its keep — and to keep moving whenever the next one arrives.
Frequently asked questions
When did GPT-5.6 become publicly available?
OpenAI made GPT-5.6 — the Sol, Terra and Luna variants — publicly available on Thursday, 9 July 2026, across ChatGPT and the API. The launch followed a roughly two-week gated preview that began in late June and was limited to a small group of vetted partner organisations at the US government's request, after which access was expanded globally.
What are the GPT-5.6 tiers Sol, Terra and Luna, and what do they cost?
GPT-5.6 ships as three tiers. Sol is the flagship at a reported $5 per million input tokens and $30 per million output. Terra is the everyday tier at $2.50 input and $15 output, described as roughly matching the prior generation at about half the cost. Luna is the lowest-cost tier at $1 input and $6 output. Rates move often, so confirm with the provider before committing a budget.
Why did GPT-5.6 go through a US government review before launch?
An AI cybersecurity order signed in early June 2026 asked companies to voluntarily submit their most powerful models to the US government roughly 30 days before public release. OpenAI ran a limited preview with vetted partners while the Department of Commerce's Center for AI Standards and Innovation conducted additional testing, then received permission for a wider release. OpenAI stated it does not believe this kind of government access process should become the long-term default.
What does the pre-release review precedent mean for teams building on frontier models?
It signals that frontier-model availability is becoming a governed, potentially delayed event rather than an instant switch. Teams should plan for release timing they do not control, keep a provider-agnostic abstraction and a configured fallback model so a gated or delayed launch does not block a roadmap, and track model provenance and testing status as part of vendor-risk and compliance records — especially in regulated verticals such as FinTech and HealthTech.
Which GPT-5.6 tier should teams use in production?
Match the tier to the task rather than defaulting to the flagship. Route hard planning, ambiguous reasoning and complex code to Sol; send everyday reasoning, drafting and tool orchestration to Terra; and push classification, extraction, tagging and formatting to Luna or an open-weight model. With a 5× spread between the top and bottom tiers, treating tier selection as an explicit routing decision — behind an abstraction, with per-step cost and quality instrumentation — is the difference between a predictable bill and a surprise.
Sources
Engadget — OpenAI gets permission to roll out GPT-5.6 to the public on July 9 (8 July 2026)
9to5Mac — OpenAI shares update on GPT-5.6 availability after holding back release (8 July 2026)