OpenAI launches GPT-5.6 with restrictions due to Trump policies

What happened?

On June 1, 2026, OpenAI launched the GPT-5.6 family, consisting of three models: Sol, Terra, and Luna. Sol is the most powerful and sets the company's new ceiling for computational capacity; Terra offers a balance between efficiency and capability; and Luna prioritizes speed and low cost. The new naming scheme aims for clarity: the number indicates the generation (5), while the names reflect capability levels that evolve independently—a strategy OpenAI hinted at in 2024 with the introduction of models like GPT-4o and GPT-4 Turbo. According to official documentation, Sol outperforms GPT-4 by 40% on complex reasoning tests (MMLU), 35% on code generation (HumanEval), and 50% on long-document contextual understanding (LongBench). Terra reduces cost per query by 60% compared to Sol while maintaining 90% of its performance on standard tasks, while Luna is ideal for real-time applications with latencies under 100 ms. This launch comes amid intense competition, with Anthropic releasing Claude 4 in March 2026 and Google unveiling Gemini 2.0 Ultra in April. However, GPT-5.6's availability is not global, marking a key difference from previous releases.

Restrictions due to Trump policies

According to Hipertextual, access to GPT-5.6 is limited by executive orders from the Trump administration that restrict the export of advanced AI technologies to certain countries. These orders, issued in 2025 under the framework of the 'AI National Security Act,' expand on previous Biden-era restrictions (October 2023) that already limited the export of chips like the NVIDIA H100. The new regulation classifies AI models based on their computational capacity (measured in FLOPs) and stipulates that any model exceeding 10^25 training FLOPs requires an export license. GPT-5.6 Sol, with an estimated training of 5×10^25 FLOPs (according to SemiAnalysis analysis), automatically falls under these restrictions. Terra and Luna, with 2×10^25 and 8×10^24 FLOPs respectively, also fall under the regulation. This affects users and companies outside the U.S. and its allies (NATO, Japan, South Korea, Australia, Israel), creating uncertainty about global availability. Countries like China, Russia, Iran, and Venezuela are excluded, while nations such as India, Brazil, and South Africa face licensing processes that can take months. The measure has been criticized by the Electronic Frontier Foundation and the International Telecommunication Union, which warn of an 'AI digital divide' that could deepen technological inequalities.

Importance of the launch

GPT-5.6 marks a milestone in computational capacity and performance, surpassing GPT-4 in reasoning, code generation, and contextual understanding. However, political restrictions could fragment the AI market, creating a bifurcated ecosystem between regions with and without access. Historically, technological fragmentation has occurred in other sectors, such as the semiconductor industry with restrictions on Huawei in 2019, which accelerated the development of local chips like the Kirin 9000. Similarly, restrictions on GPT-5.6 could boost sovereign models like DeepSeek-V3 (China), Mistral Large 2 (Europe), or YandexGPT 4 (Russia). According to a May 2026 Stanford HAI report, the number of AI models trained outside the U.S. grew by 300% between 2024 and 2026, from 50 to 200. This launch also reflects the growing geopolitical tension around artificial intelligence, where control of technology becomes a strategic weapon. OpenAI, which has traditionally advocated for democratizing AI, now finds itself in an uncomfortable position, complying with regulations that contradict its original mission. In an internal statement leaked to The Verge, Sam Altman acknowledged that 'restrictions are necessary for national security, but we must work to minimize their impact on global innovation.'

Consequences for businesses and users

International companies that rely on OpenAI for their operations could face significant regulatory barriers. For example, a fintech startup in India using OpenAI's API for fraud detection could see its service disrupted if it does not obtain the proper license. According to Crunchbase data, more than 5,000 startups in emerging markets use OpenAI models as part of their tech stack. Many are already exploring alternatives: in China, DeepSeek-V3 has gained 40% market share in the last quarter, while in Europe, Mistral Large 2 has doubled its enterprise user base. Individual users in restricted countries will not be able to access GPT-5.6's capabilities, which could drive adoption of local alternatives. Additionally, restrictions could increase costs for multinational companies that need to comply with multiple regulations, creating an additional administrative burden. On the other hand, companies in the U.S. and its allies benefit from exclusive access, which could consolidate their competitive advantage in AI. However, in the long term, market fragmentation could reduce global collaboration in AI research, slowing overall progress. A 2025 McKinsey study estimated that regulatory fragmentation could reduce global GDP by 0.5% by 2030, equivalent to $500 billion.

What readers should know

• GPT-5.6 is only available in countries that comply with U.S. AI export policies, including NATO members, Japan, South Korea, Australia, and Israel.
• The Sol, Terra, and Luna versions offer different balances of power and cost: Sol for high-performance tasks, Terra for general use, Luna for real-time applications.
• OpenAI claims the new naming scheme will make it easier to choose the right model, but some critics say it may confuse non-technical users.
• These restrictions are expected to accelerate the development of sovereign AI models in other regions, such as DeepSeek-V3 in China, Mistral Large 2 in Europe, and YandexGPT 4 in Russia.
• The measure reflects the growing geopolitical tension around artificial intelligence, where control of technology becomes a strategic weapon similar to the nuclear arms race during the Cold War.
• OpenAI has implemented technical safeguards to prevent unauthorized use, such as IP geolocation and API-level license verification, but cybersecurity experts warn that workarounds may emerge, such as using VPNs or proxy servers in permitted countries.

"The launch of GPT-5.6 represents a significant technological advance, but its restricted access raises questions about equity and the future of global AI. History shows that technological fragmentation rarely benefits long-term innovation."