On 26/06/2026, OpenAI announced the GPT-5.6 model family with three versions: Sol (flagship), Terra (mid-tier) and Luna (fast/cheap). But the most striking part isn't the benchmarks — it's the release mechanism: at the US government's request, in the initial phase only about 20 identity-vetted "trusted partners" can access Sol. This is the first time a US frontier model has launched under a government-managed access list.
Update 24/09/2026: the restricted phase is over. OpenAI released the GPT-5.6 family, Sol included, for general availability on 09/07/2026 (per the OpenAI API changelog and the GPT-5.6 announcement). On 21/08/2026 Sol's API price was cut to $4 input / $20 output per 1M tokens, a promotional price available at least through 21/11/2026 (per the GPT-5.6 Sol model page); OpenAI had earlier cut Luna by 80% and Terra by 20% on 30/07. The prices in the tables below are the launch prices.
Quick summary
- When: preview 26/06/2026; the vetting mechanism follows the AI executive order signed by President Trump on 02/06.
- 3 models: Sol (flagship) · Terra (mid-tier) · Luna (fast/cheap).
- Launch price per 1M tokens: Sol $5/$30 · Terra $2.50/$15 · Luna $1/$6 (Sol later cut to $4/$20).
- Sol: completes ~50% of long-running expert tasks; strongest in coding, biology, cybersecurity; has a 'max reasoning effort' mode.
- OpenAI says: this vetting mechanism 'shouldn't become the long-term default'; the family, Sol included, reached general availability on 09/07/2026.
What happened?
According to OpenAI and major outlets (TechCrunch, Forbes, CNBC), GPT-5.6 Sol is the company's most powerful model to date — but initial access is limited to ~20 organizations identity-vetted by the US government. Terra and Luna are more widely available, while Sol must wait for the approval process.
| Version | Positioning | Input / 1M | Output / 1M |
|---|---|---|---|
| Sol | Flagship | $5 | $30 |
| Terra | Mid-tier | $2.50 | $15 |
| Luna | Fast/cheap | $1 | $6 |
This mechanism is a direct consequence of the AI executive order signed on 02/06 — the same policy framework that led to Fable 5 being forced offline worldwide in mid-June.
Why this is a notable precedent
Within a single month, the US AI industry has seen two unprecedented control mechanisms: a live model taken offline by order (Fable 5), and now a new model launching under an approved-access list (GPT-5.6 Sol). Add the fact that Mythos 5 was reopened to only ~100 selected organizations, and the trend is clear: access to the most powerful AI is being tiered by trust levels defined by the state.
OpenAI has publicly said this mechanism "shouldn't become the long-term default" — at the time, users outside the list (including most businesses outside the US) could only wait. In the end the wait lasted about two weeks: GPT-5.6 reached general availability on 09/07/2026. The precedent remains: the release timing of the most powerful models can depend on a government request.
The Namtech angle for Vietnamese businesses
For Vietnamese businesses, the message is increasingly clear: the path to the most powerful models no longer depends on budget alone — it also depends on the policy of the country where the model was created. Today it's a list of 20 partners; tomorrow it could be a different set of criteria.
The sensible strategy: leverage cloud models when permitted and appropriate, but build an internal AI foundation layer on open-source models — something no one can revoke with an administrative decision. That is the philosophy behind Namtech's private internal AI platform.
Frequently asked questions
Which models are in GPT-5.6?
Three versions: Sol (flagship), Terra (mid-tier), Luna (fast/cheap). Launch price per 1M tokens: Sol $5/$30, Terra $2.50/$15, Luna $1/$6; Sol was cut to $4/$20 on 21/08/2026 and Luna and Terra were cut on 30/07/2026.
Who can use GPT-5.6 Sol?
In the initial phase (from 26/06/2026), only about 20 'trusted partners' identity-vetted by the US government. Since 09/07/2026 GPT-5.6, Sol included, has been generally available across ChatGPT, Codex and the OpenAI API.
Why did the US government step in?
The vetting mechanism follows the AI executive order signed on 02/06/2026 — the same policy framework that led to the forced takedown of Anthropic's Fable 5.
Are Vietnamese businesses affected?
Quite possibly: when access to the most powerful models is allocated via a US government approval list, businesses outside the list must wait or use weaker versions. This is why an independent internal AI layer matters.
Don't wait in line for approval
Namtech deploys internal AI on open-source models — running on your infrastructure, independent of anyone's access list.
Book a free consultationNote: This article was first compiled from public sources on 02/07/2026. Last updated 24/09/2026: general availability on 09/07/2026 and later price cuts added. For reference only, subject to change.