OpenAI will publicly launch GPT-5.6, its most advanced artificial intelligence model to date, on Thursday after delaying the release last month following requests from the U.S. government for additional national security reviews amid growing concerns over the potential misuse of increasingly powerful AI systems.
The launch comes as competition between the United States and China to develop next-generation AI models intensifies, with governments on both sides imposing tighter oversight on frontier AI technologies that could have significant military, intelligence and cybersecurity applications.
According to Axios, which first reported the launch, the Trump administration approved the broader release of GPT-5.6 after OpenAI completed additional safety testing and held a series of meetings with U.S. government officials.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
U.S. officials have expanded scrutiny of advanced AI models over concerns that highly capable systems could accelerate sophisticated cyberattacks, automate vulnerability discovery, assist biological research or enhance military and intelligence operations if they become widely accessible to geopolitical rivals.
Security experts have warned that increasingly capable AI systems could help attackers identify weaknesses in critical infrastructure, including power grids, financial networks, telecommunications systems and industrial control systems that often rely on complex software and legacy technologies.
Washington has therefore intensified reviews of frontier AI releases to evaluate whether new models could pose unacceptable risks if exploited by hostile governments or state-backed cyber groups in countries such as China and Russia.
China has taken similar precautions.
Chinese regulators have reportedly held discussions with leading domestic technology companies over restricting overseas access to China’s most advanced AI models, including systems that have yet to be publicly released, reflecting growing concerns that frontier AI capabilities have become strategically important national assets.
During the review period, OpenAI limited GPT-5.6 access to a small group of vetted partners whose identities were shared with U.S. authorities. The restricted rollout enabled the company and government agencies to conduct additional evaluations before approving a wider release.
OpenAI announced in a post on X late Tuesday that it will introduce three new models to the public:
- GPT-5.6 Sol, its flagship and most capable model.
- GPT-5.6 Terra, a lower-cost alternative.
- GPT-5.6 Luna, designed to provide more affordable access for developers and businesses.
In the AI industry, companies are adopting a multi-model strategy, increasingly offering premium, mid-tier, and lightweight models to serve different performance requirements and computing budgets.
When OpenAI previewed GPT-5.6 in late June, it highlighted significant improvements in so-called agentic capabilities, enabling the model to perform more complex multi-step tasks with greater autonomy. The company said the new generation delivers stronger performance across software engineering, biological research, and cybersecurity.
OpenAI also reported that GPT-5.6 Sol performed competitively against Anthropic’s Mythos Preview on ExploitBench, a benchmark designed to evaluate AI models’ cybersecurity capabilities.
Performance in cybersecurity has become one of the most closely watched measures for frontier AI because advances in offensive and defensive cyber capabilities could carry significant implications for governments, corporations and critical infrastructure operators.
Anthropic And Musk Also Expand AI Offerings
The GPT-5.6 launch follows a series of major announcements across the AI industry. OpenAI rival Anthropic temporarily disabled public access to its most advanced models, Mythos 5 and Fable 5, after the U.S. government’s June 12 export control order raised national security concerns about advanced AI technologies.
The restrictions were lifted last week after Anthropic implemented additional safeguards requested by U.S. authorities.
Meanwhile, billionaire Elon Musk said on Wednesday that his AI company, SpaceXAI, would also make its flagship Grok 4.5 model publicly available, further intensifying competition among leading AI developers.
The rapid succession of releases highlights the competitive race among OpenAI, Anthropic, Google, xAI, Meta Platforms, and Chinese AI developers to establish leadership in frontier model performance.
OpenAI Expands Into Real-Time Voice AI
Alongside GPT-5.6, OpenAI is also broadening its portfolio of multimodal AI products. On Wednesday, the company launched GPT-Live, a new family of real-time voice models capable of listening and speaking simultaneously, allowing conversations to flow more naturally without waiting for users to finish speaking.
The company said two versions, GPT-Live-1 and GPT-Live-1 mini, will be rolled out globally.
The models build on OpenAI’s strategy of making AI assistants more conversational and responsive, enabling them to support customer service, productivity tools, education and enterprise applications that require continuous spoken interaction.
The launch follows OpenAI’s introduction in May of three new audio models for developers, designed to make voice-based AI agents more natural and capable of completing tasks in real time.