OpenAI is officially releasing ChatGPT 5.6 into broad public access on July 9, 2026 — after an almost two-week delay related to the requirements of the U.S. administration for a preliminary government review. We explain what kind of model this is, how much it costs, and why its release turned into a political story.
When ChatGPT 5.6 Came Out: Launch Timeline
The first version of ChatGPT 5.6 appeared back on June 26, 2026, but in a limited format: access was granted only to a small group of verified partners whose participation had been coordinated with the U.S. government. OpenAI stated directly that it did not consider this practice a long-term norm, but agreed to it as the fastest path to a full release.
On Thursday, July 9, the company opened access to all three versions of the model — Sol, Terra and Luna — for all users around the world. At the same time, OpenAI introduced a new generation of voice models called GPT-Live.
Why the Release Was Delayed: White House Intervention
The reason for the delay was an executive order signed by President Trump in early June 2026. The order requires developers of the most powerful AI models to voluntarily submit their systems for government evaluation 30 days before public release, while giving federal agencies 60 days to develop the evaluation procedure itself (deadline — August 1, 2026).
Because of this, OpenAI initially limited access to ChatGPT 5.6 to verified partners only. The relevant division of the U.S. Department of Commerce — the Center for AI Standards and Innovation — conducted additional testing of the model, while the company sent its own technical specialists to Washington to provide rapid answers to regulators’ questions.
In the end, OpenAI did not have to wait the full 30 days: the broad release happened after just two weeks. However, there is an important nuance: a White House representative officially stated that the administration had not given OpenAI any formal “permission” for the release, emphasizing that the decision about the timing and scale of the model’s launch remains entirely with the company. This statement came in response to publications that described the situation as direct government approval.
Parallel With Claude Fable 5
This is already the second similar case in a month. Anthropic’s Claude Fable 5 model was temporarily blocked for international users due to U.S. export restrictions, and access was restored only on July 1, 2026 — after almost three weeks of downtime. Both episodes show how closely government policy is now intertwined with AI model release dates.
Three Versions of ChatGPT 5.6: Sol, Terra, Luna
OpenAI has moved away from a single model in favor of three versions with different levels of power and pricing — an approach similar to what competitors have previously used.
Sol — the Flagship Model
Sol is the most powerful version of the lineup, aimed at complex tasks: coding, scientific research and cybersecurity. According to OpenAI’s internal benchmarks, Sol scores 88.8% on TerminalBench 2.1, while the expanded Sol Ultra version scores 91.9%, higher than Claude Mythos 5’s result (88.0%) on the same test.
Terra — Balance of Price and Quality
Terra is positioned as a model for everyday use: according to the company, it delivers performance on the level of GPT-5.5 while costing half as much. On TerminalBench 2.1, Terra scored 84.3%, surpassing its previous version.
Luna — the Budget Version
Luna is the most mass-market and cheapest model in the lineup. It scored 82.5% on TerminalBench 2.1 — higher than Claude Opus 4.8 (78.9%), but lower than GPT-5.5 (83.4%).
How Much ChatGPT 5.6 Costs
Prices are listed per million tokens (input / output):
- Sol — $5 / $30
- Terra — $2.50 / $15
- Luna — $1 / $6
An interesting point for comparison: on July 7, Anthropic changed the pricing model for Claude Fable 5 — the model moved from a subscription to a paid-credit usage system and now costs $10 per million input tokens and $50 per million output tokens. This means that OpenAI’s top model (Sol) is now exactly twice as cheap as Anthropic’s top model — at least according to official pricing at the time of publication.
Cybersecurity: Why This Release Is Special
According to OpenAI, Sol is the company’s most powerful model from a cybersecurity standpoint: it demonstrates substantial progress in long-session vulnerability discovery tasks, showing results on the level of Anthropic’s Mythos Preview model while using roughly three times fewer tokens.
It is important to understand the limits of this capability: according to the official system card, Sol did not cross the “Critical” cyber-risk threshold under the company’s internal Preparedness Framework. In tests on real web browsers, the model found individual vulnerabilities and components from which an exploit could theoretically be assembled, but it was unable to independently create a full working attack chain under test conditions.
Because of this growth in capabilities, OpenAI applied its strictest set of safeguards to date: more than 700,000 hours of compute on A100e GPUs were allocated to automated searches for universal jailbreaks, and the system is undergoing continuous “red teaming” — testing for resistance to real attacks — throughout the entire deployment period.
GPT-Live Voice Models
Alongside ChatGPT 5.6, the company introduced a new generation of voice models called GPT-Live (previously known internally as GPT-Bidi 1). Details of the capabilities have not yet been fully disclosed — we expect additional information in the coming days and will update the article.
In Brief: The Key Facts About ChatGPT 5.6
- Broad release date — July 9, 2026 (Sol, Terra, Luna simultaneously)
- First limited version — June 26, 2026, only for verified partners
- Reason for the delay — an executive order on preliminary government review of frontier models
- Important: the White House did not officially “approve” the release — the decision remained with OpenAI
- Prices: Sol $5/$30, Terra $2.50/$15, Luna $1/$6 per million tokens
- Comparison: Sol is twice as cheap as the updated Claude Fable 5 price ($10/$50)
- TerminalBench 2.1 benchmark: Sol Ultra 91.9% — higher than Claude Mythos 5 (88.0%)
- Sol did not cross the “Critical” cyber-risk threshold, but received the strictest safeguards in the company’s history
FAQ
When did ChatGPT 5.6 come out? The limited version appeared on June 26, 2026, only for verified partners. Broad public access to all three versions (Sol, Terra, Luna) opened on July 9, 2026.
Why was the ChatGPT 5.6 release delayed? Because of an executive order by the Trump administration requiring developers of the most powerful AI models to undergo a preliminary government review 30 days before public release.
How much does ChatGPT 5.6 cost? Sol costs $5 per million input tokens and $30 per million output tokens. Terra — $2.50/$15. Luna — $1/$6.
How are Sol, Terra and Luna different? Sol is the most powerful version for complex tasks (coding, research, cybersecurity). Terra balances price and quality for everyday use. Luna is the cheapest, mass-market version.
Did the White House approve the ChatGPT 5.6 release? Officially, no. An administration representative directly stated that no formal permission had been granted, and that the release timing decision remained with OpenAI.
What is GPT-Live? A new generation of OpenAI voice models introduced alongside ChatGPT 5.6. Details of the capabilities are still being clarified.
The article was prepared by the TechVisor team — practical IT media for people.
