![\"Why](\"https:\/\/techvisor.pro\/wp-content\/uploads\/2026\/07\/chomu-reliz-zatrymaly.webp\" "\\"\\"")
<\/p>\nOpenAI is officially releasing ChatGPT 5.6 into broad public access on July 9, 2026 \u2014 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.<\/p>\n
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.<\/p>\n
On Thursday, July 9, the company opened access to all three versions of the model \u2014 Sol, Terra and Luna \u2014 for all users around the world. At the same time, OpenAI introduced a new generation of voice models called GPT-Live.<\/p>\n
<\/p>\n
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 \u2014 August 1, 2026).<\/p>\n
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 \u2014 the Center for AI Standards and Innovation \u2014 conducted additional testing of the model, while the company sent its own technical specialists to Washington to provide rapid answers to regulators\u2019 questions.<\/p>\n
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 \u201cpermission\u201d for the release, emphasizing that the decision about the timing and scale of the model\u2019s launch remains entirely with the company. This statement came in response to publications that described the situation as direct government approval.<\/p>\n
This is already the second similar case in a month. Anthropic\u2019s 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 \u2014 after almost three weeks of downtime. Both episodes show how closely government policy is now intertwined with AI model release dates.<\/p>\n
![\"Three](\"https:\/\/techvisor.pro\/wp-content\/uploads\/2026\/07\/try-versiyi-chatgpt-5.6.webp\" "\\"\\"")
<\/p>\n
OpenAI has moved away from a single model in favor of three versions with different levels of power and pricing \u2014 an approach similar to what competitors have previously used.<\/p>\n
Sol is the most powerful version of the lineup, aimed at complex tasks: coding, scientific research and cybersecurity. According to OpenAI\u2019s internal benchmarks, Sol scores 88.8% on TerminalBench 2.1, while the expanded Sol Ultra version scores 91.9%, higher than Claude Mythos 5\u2019s result (88.0%) on the same test.<\/p>\n
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.<\/p>\n
Luna is the most mass-market and cheapest model in the lineup. It scored 82.5% on TerminalBench 2.1 \u2014 higher than Claude Opus 4.8 (78.9%), but lower than GPT-5.5 (83.4%).<\/p>\n
Prices are listed per million tokens (input \/ output):<\/p>\n
An interesting point for comparison: on July 7, Anthropic changed the pricing model for Claude Fable 5 \u2014 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\u2019s top model (Sol) is now exactly twice as cheap as Anthropic\u2019s top model \u2014 at least according to official pricing at the time of publication.<\/p>\n
![\"Cybersecurity\"](\"https:\/\/techvisor.pro\/wp-content\/uploads\/2026\/07\/kiberbezpeka.webp\" "\\"\\"")
<\/p>\n
According to OpenAI, Sol is the company\u2019s most powerful model from a cybersecurity standpoint: it demonstrates substantial progress in long-session vulnerability discovery tasks, showing results on the level of Anthropic\u2019s Mythos Preview model while using roughly three times fewer tokens.<\/p>\n
It is important to understand the limits of this capability: according to the official system card, Sol did not cross<\/strong> the \u201cCritical\u201d cyber-risk threshold under the company\u2019s 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.<\/p>\n 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 \u201cred teaming\u201d \u2014 testing for resistance to real attacks \u2014 throughout the entire deployment period.<\/p>\n 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 \u2014 we expect additional information in the coming days and will update the article.<\/p>\n When did ChatGPT 5.6 come out?<\/strong> 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.<\/p>\n Why was the ChatGPT 5.6 release delayed?<\/strong> 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.<\/p>\n How much does ChatGPT 5.6 cost?<\/strong> Sol costs $5 per million input tokens and $30 per million output tokens. Terra \u2014 $2.50\/$15. Luna \u2014 $1\/$6.<\/p>\n How are Sol, Terra and Luna different?<\/strong> 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.<\/p>\n Did the White House approve the ChatGPT 5.6 release?<\/strong> Officially, no. An administration representative directly stated that no formal permission had been granted, and that the release timing decision remained with OpenAI.<\/p>\n What is GPT-Live?<\/strong> A new generation of OpenAI voice models introduced alongside ChatGPT 5.6. Details of the capabilities are still being clarified.<\/p>\n The article was prepared by the TechVisor team \u2014 practical IT media for people.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":" OpenAI is officially releasing ChatGPT 5.6 into broad public access on July 9, 2026 \u2014 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 […]<\/p>\n","protected":false},"author":4,"featured_media":2072,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[21,25],"tags":[],"class_list":["post-2073","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news","category-ai-hub"],"blocksy_meta":[],"_links":{"self":[{"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/posts\/2073","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/comments?post=2073"}],"version-history":[{"count":1,"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/posts\/2073\/revisions"}],"predecessor-version":[{"id":2079,"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/posts\/2073\/revisions\/2079"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/media\/2072"}],"wp:attachment":[{"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/media?parent=2073"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/categories?post=2073"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techvisor.pro\/en\/wp-json\/wp\/v2\/tags?post=2073"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}GPT-Live Voice Models<\/h2>\n
In Brief: The Key Facts About ChatGPT 5.6<\/h2>\n
\n
FAQ<\/h2>\n
\n