GPT-5.6 Sol, Terra, Luna Launch With Government-Vetted Access List Before Wider Rollout

GigaNectar Team

Abstract generative art graphic accompanying OpenAI's GPT-5.6 announcement
Interactive Access Tracker

Who Gets To Use OpenAI’s Strongest Model First, And Why The White House Decides

GPT‑5.6 Sol, Terra, and Luna are live, but only a small, government‑shared list of organizations can reach them right now. Trace the access pathway below, then open each panel to see what changes between tiers, risk levels, and price.

01 / LAB OpenAI finishes Sol, Terra, Luna
02 / PREVIEW Capabilities shared with the U.S. government
03 / VETTING Partner list reviewed under June 2 order
04 / ACCESS API and Codex open to cleared partners
05 / AHEAD Broader rollout planned in coming weeks
Tier Breakdown

Three Models, Three Jobs: Compare Sol, Terra, And Luna

Tap a tier to see its pricing, its standout benchmark, and what kind of task OpenAI built it to handle.

Input / Output (per 1M tokens)
$5 / $30
Role
Flagship
Headline Benchmark
Terminal‑Bench 2.1 state of the art
New Modes
Max reasoning + Ultra subagents
Sol is the version built for the hardest, longest jobs — multi-step coding work, genomics analysis, and vulnerability research. It is the only tier that unlocks Ultra mode, where subagents split a task and work on pieces of it at the same time.
Input / Output (per 1M tokens)
$2.50 / $15
Role
Everyday work
Positioning
Matches GPT‑5.5, 2x cheaper
Best Fit
High‑volume business tasks
Terra is OpenAI’s middle tier: support tickets, document review, and internal tools at a scale where cost per request matters as much as raw capability. OpenAI says it holds up against GPT‑5.5 while costing half as much.
Input / Output (per 1M tokens)
$1 / $6
Role
Fast and affordable
Positioning
Lowest cost in the family
Best Fit
Summaries, routing, routine drafts
Luna is the volume tier: quick replies, classification, and everyday drafting where speed and price matter more than squeezing out the last bit of reasoning depth.
Rows of server racks in a data center corridor lit by blue light
Caption: Server infrastructure of this kind underpins the inference capacity OpenAI is allocating to a limited group of preview partners before wider GPT‑5.6 access.
Photo Source: Unsplash (Free to use under the Unsplash License)
Alt Text: Rows of server racks in a data center corridor lit by blue light
Preparedness Framework

Why Every Tier In The Family Got Flagged As High Risk

OpenAI’s own system card rates the cyber and biology risk of the family. Here is how that compares with the one threshold none of them crossed.

Cybersecurity Capability
LowHighCritical
Sol, Terra, and Luna are all rated High under OpenAI’s Preparedness Framework for cybersecurity, the first time smaller, faster tiers in a family have reached that rating alongside the flagship.
Biological / Chemical Risk
LowHighCritical
All three models are also rated High for biological and chemical risk, a separate tracked category in the same framework.
Critical Threshold (Not Reached)
LowHighCritical
In testing against Chromium and Firefox, Sol found exploitation primitives but did not independently build a complete, working exploit. That keeps it under the framework’s Critical line.
Regulatory Trail

How One June Order Set The Rules For This Launch

Tap each entry to expand the detail behind it.

President Trump signs the AI cybersecurity orderJUN 2, 2026

The White House issued “Promoting Advanced Artificial Intelligence Innovation and Security,” directing federal agencies to build a voluntary framework letting AI developers share advanced models with the government for up to 30 days before release to trusted partners. The order also creates a classified process for deciding when a model counts as a “covered frontier model,” a call made by the Director of the National Security Agency.

Agencies given a 60-day deadline to build the review processBY AUG 1, 2026

The order tasks the Treasury, War, and Homeland Security departments, working with the NSA, CISA, and NIST, with finishing the classified benchmarking process and the voluntary early-access framework within 60 days. That process did not exist yet when GPT‑5.6 launched, which OpenAI says is why it agreed to a phased rollout in the meantime.

OpenAI previews GPT‑5.6 Sol, Terra, and LunaJUN 26, 2026

OpenAI opened a limited preview through the API and Codex only, describing it as the company’s strongest model family yet and pairing the release with what it calls its most robust safety stack to date. ChatGPT access was not included in this preview.

Access starts with a government-shared partner listJUN 26, 2026

OpenAI says it shared its release plans and the models’ capabilities with the U.S. government ahead of launch and, at the government’s request, is starting with a small group of trusted partners and organizations whose participation has been shared with officials.

Broader rollout planned for the coming weeksAHEAD

OpenAI says it intends to make Sol, Terra, and Luna generally available across ChatGPT, Codex, and the API soon, while continuing to test and coordinate with partners during the preview period. No fixed general-availability date has been announced.

Close-up of a circuit board with glowing data pathways
Caption: The layered safeguard system behind GPT‑5.6 runs checks during generation as well as on the model’s training, a structure OpenAI compares to several independent checkpoints rather than one single gate.
Photo Source: Unsplash (Free to use under the Unsplash License)
Alt Text: Close-up of a circuit board with glowing data pathways
Token Economics

What Cleared Partners Are Actually Paying

Sol Input
$5
per 1M tokens
Sol Output
$30
per 1M tokens
Terra Input
$2.50
per 1M tokens
Terra Output
$15
per 1M tokens
Luna Input
$1
per 1M tokens
Luna Output
$6
per 1M tokens

GPT‑5.6 also introduces explicit prompt-cache breakpoints with a 30-minute minimum cache life. Cache writes cost 1.25x the model’s standard input rate, while cache reads keep the existing 90 percent discount.

Leave a comment