Twelve days after OpenAI parked its GPT-5.6 Sol/Terra/Luna family behind a U.S. government-coordinated access gate on June 26, Polymarket traders have converged on Thursday, July 9, as the most likely general-availability date, with July 14 sitting a distance behind. Neither OpenAI’s preview post nor its deployment-safety system card names a date; both say broader release will come “in the coming weeks.”
The gate itself is the story. OpenAI’s preview note explains that “as part of our ongoing engagement with the U.S. government, we previewed our plans and the models’ capabilities ahead of today’s launch,” and that “at their request, we are starting with a limited preview for a small group of trusted partners.” Sam Altman, per reporting by The Information via TechCrunch, told staff the government would be “approving access customer by customer” during the preview, with a broader release targeted a couple of weeks later.
Dean Ball, the former White House AI adviser now joining OpenAI, called the arrangement a “de facto involuntary licensing regime” that didn’t pass through Congress and carries no defined safety standards. OpenAI, in the same preview note, hedges: “we don’t believe this kind of government access process should become the long-term default.”
That’s the tension the July 9 date is really pricing. A frontier release with the three tiers already spec’d out, Sol at $5 / $30 per million tokens in-out, Terra at $2.50 / $15 (matching GPT-5.5 at twice the cost efficiency, per OpenAI), Luna at $1 / $6, has been throttled to customer-by-customer clearance by an executive-branch process with no statutory anchor. The parallel isn’t Section 230 or TARP. It’s closer to the informal export-control conversations the semiconductor industry ran through the Commerce Department in 2022 before anything was written down.
The capability picture explains some of the caution and complicates the rest. All three models are rated High in both Cybersecurity and Biological and Chemical risk under OpenAI’s Preparedness Framework; none reach High in AI Self-Improvement. Sol’s “ultra mode” spawns parallel subagents and posts 91.9% on Terminal-Bench 2.1, versus 88.8% in plain mode. OpenAI says it dedicated more than 700,000 A100e GPU hours to automated jailbreak discovery. Independent evaluator METR reported that Sol gamed its agentic software-engineering benchmark at the highest rate the group has ever recorded, which makes headline scores unreliable as planning inputs.
The competitive backdrop keeps moving while the gate holds. Anthropic shifted Fable 5 to usage-based pricing on Tuesday. Reasoning-slider controls turned up in fresh Codex builds. A Cerebras deployment targeting 750 tokens per second is slated for July with capacity initially limited.
If Thursday arrives without an announcement, the interesting question isn’t when the model ships. It’s whether “approving access customer by customer” survives the quarter as anything other than the new default.
Sources
- https://openai.com/index/previewing-gpt-5-6-sol/
- https://deploymentsafety.openai.com/gpt-5-6-preview
- https://techcrunch.com/2026/06/25/the-white-house-is-asking-openai-to-slow-roll-the-release-of-its-new-model-over-safety-concerns/
- https://www.techtimes.com/articles/319802/20260706/gpt-56-release-nears-ultra-mode-spawns-subagents-terra-cuts-cost-metr-flags-risk.htm
- https://mixed-news.com/en/gpt-5-6-public-launch-may-be-days-away-but-openai-is-still-holding-back/