OpenAI made it in less than a day. On Wednesday evening, news leaked that the Trump administration was requesting a delay for GPT-5.6 – and by Friday, the new model was already here. OpenAI unveiled a limited preview of a trio of models: Sol, Terra, and Luna. However, the premiere was not without political drama: the US government will approve customer access on a case-by-case basis. We have not seen anything like this in the history of commercial AI. Three models for three different needs With GPT-5.6, OpenAI is for the first time introducing an entire family of models at once – each targeting a different type of user. Sol is the flagship: the most powerful model for demanding tasks such as programming, cybersecurity, or biological research. Terra is the mid-tier designed for "high-volume work" – suitable for companies that need large-scale performance. And Luna is a lightweight model for everyday use, designed to be "fast and affordable". All three models excel particularly in long-term agent tasks – i.e., situations where AI works autonomously for an extended period without the need for constant human guidance. This is a key shift from GPT-5.5, which, while offering better reasoning, occasionally lost track of longer tasks. OpenAI claims that GPT-5.6 will maintain context and focus even with complex, hour-long workflows. Prices that make sense – and pressure Anthropic GPT-5.6's pricing policy is another surprise. The flagship Sol costs 5 dollars per million input tokens and 30 dollars per million output tokens. This is almost half the price of Anthropic's Claude Fable 5, which charges 10 dollars for input and 50 for output. Terra is half the price of Sol, and Luna is less than half the price of Terra. For context: a million tokens equals approximately 750,000 words – several thick novels. For the average ChatGPT user, nothing changes; the models are available within the existing subscription. However, companies and developers using the API will experience significant savings compared to the competition. OpenAI is thus sending a clear signal: it is serious about the price war with Anthropic. And this is despite the company recently admitting that current prices are not sustainable in the long term. Max and Ultra: two new modes that evoke OpenClaw Sol comes with two special modes. Max Mode provides deeper reasoning for the most complex tasks – for example, mathematical proofs or detailed code security audits. Ultra Mode goes even further: it uses sub-agents, i.e., "subordinate" AI instances that solve partial problems in parallel. The concept of sub-agents strongly resembles OpenClaw – a system created by Peter Steinberger, who now works at OpenAI. It seems Steinberger's work is already bearing its first fruits. Safety first – perhaps too much OpenAI dedicated most of its announcement blog post to safety. The company stated that it spent approximately 700,000 A100e GPU hours on automated red-teaming (simulated attacks on the model) and collaborated with external testers who will continue to audit the model for another two weeks. The model is "trained to reject prohibited cyber assistance, including instances where users attempt to mask their intent or jailbreak the model." This is an obvious allusion to Anthropic's recent problems, whose Mythos and Fable models faced criticism for their exploitability. However, OpenAI admits that safety features can sometimes interfere with legitimate work – especially in areas where defensive and offensive activities initially resemble each other. "That's exactly what the preview mode is designed to test," the company wrote. The Trump administration changes the rules of the game The launch of GPT-5.6 takes place against the backdrop of the biggest regulatory crisis in the history of commercial AI. In early June, the Trump administration ordered Anthropic to suspend access to Mythos 5 and Fable 5 models and prohibited access to them for all "foreign nationals" – including Anthropic's own employees without US citizenship. This caused an uproar in the tech sector. OpenAI fared significantly better. Instead of a blanket ban, it received a case-by-case approval regime. During the preview period, the Trump administration will personally approve every customer who gains access to the model. However, OpenAI indicates that it is not satisfied with this situation: "We do not believe this kind of government access process should become a long-term standard. It keeps the best tools from the users, developers, companies, cyber defenders, and global partners who need them." The company is also working with the administration to develop an executive order on AI cybersecurity and a "repeatable process for future model releases." General availability of GPT-5.6 is expected in a matter of weeks. What this means for Czechia and Europe For Czech companies and developers, GPT-5.6 brings both good and bad news. Good: lower API prices mean more accessible AI for smaller teams and startups. Bad: the US government approval process may complicate or delay access for European customers. From the perspective of the EU AI Act, the situation is paradoxical. The European Union is creating its own regulatory framework, but the US government is now introducing even stricter control. It remains to be seen whether GPT-5.6 will even reach European customers during the preview period, or if we will have to wait for the full launch. Czech language support: GPT-5.6 – just like previous OpenAI models – understands Czech and communicates in Czech at a very good level, including understanding Czech realities and context. This makes it a fully-fledged tool for Czech users and companies as well. Context: Security Panic in Washington The entire situation surrounding GPT-5.6 is part of a broader security panic that has gripped Washington. Following the release of Anthropic Mythos – a model capable of finding critical software vulnerabilities on a massive scale – the US government began to act. The administration, which originally promised a "speed wins" approach and support for American AI exports, is suddenly imposing unprecedented restrictions. The uneven approach to individual companies – Anthropic received a strict ban, OpenAI a milder regime – raises questions about the fairness of regulation. Whether this is a deliberate strategy or a chaotic reaction, only time will tell. When will GPT-5.6 be available for regular ChatGPT users? OpenAI has not yet specified a concrete date. During the preview period, access is limited to approved customers. Full availability, including integration into ChatGPT, is expected "in a matter of weeks" – estimated during July 2026. How does GPT-5.6 compare to the competition? OpenAI has not yet published complete benchmark results. However, the price of the flagship Sol model (5/30 USD per million tokens) is almost half that of Anthropic's Claude Fable 5 (10/50 USD). The key advantage is expected to be its ability to maintain focus on long agent tasks – an area where GPT-5.5 sometimes lagged. Will US regulation affect the availability of GPT-5.6 in Czechia? During the preview period, it is possible that European customers may have to wait longer due to the American approval process. However, upon full launch, the model should be globally available, including in the Czech Republic. OpenAI itself emphasizes that it considers the current restrictions temporary.