OpenAI to launch GPT-5.6, its most advanced AI model yet, after delayed rollout over U.S. security review

OpenAI will publicly launch GPT-5.6 on Thursday, ending a delay prompted by U.S. government concerns about the national security risks posed by more capable AI systems.

The launch marks one of the clearest signs yet that frontier AI releases are no longer just product events. They are becoming matters of government review, export policy, cyber risk, and geopolitical competition.

“We’re beginning a limited preview of the GPT-5.6 series: Sol, our flagship model; Terra, a balanced model for everyday work; and Luna, a fast and affordable model. Terra has competitive performance to GPT-5.5 while being 2x cheaper and Luna brings strong capability at our lowest cost,” OpenAI said in a news release.

The company added: “GPT-5.6 Sol launches with our most robust safety stack to date. We strengthened protections for higher-risk activity, sensitive cyber requests, and repeated misuse, and spent multiple weeks finding weaknesses, pressure-testing our system, and hardening it against real-world attacks.”

OpenAI Releases GPT-5.6 Following U.S. National Security Review, Introduces Sol, Terra, and Luna

The GPT-5.6 lineup includes three models: Sol, Terra, and Luna. Sol is the flagship system and OpenAI’s most capable model to date. Terra is pitched as a lower-cost option for everyday work, with performance that OpenAI says is competitive with GPT-5.5. Luna is the cheapest model in the series, aimed at users who need speed and lower costs.

OpenAI said GPT-5.6 Sol brings gains in coding, biology, and cybersecurity, three areas that have drawn growing attention from policymakers and safety researchers.

Credit: OpenAI

According to OpenAI, “GPT-5.6 Sol is our strongest model yet. To preview model performance, we share a set of evaluations highlighting improved agentic capabilities in coding, biology, and cybersecurity, with additional safety and preparedness evaluations available in our system card (opens in a new window). We will share an expanded suite of evaluation results when we make the model broadly available.

With GPT-5.6, we’re introducing a new maximum reasoning effort to give Sol more time to reason deeply. Additionally, we’re introducing a new ultra mode that goes beyond the capabilities of a single agent by leveraging subagents to accelerate complex work.

For coding workflows, GPT-5.6 Sol sets a new state of the art on Terminal-Bench 2.1, which tests command-line workflows requiring planning, iteration, and tool coordination.”

TerminalBench 2.1 (Source: OpenAI)

GPT-5.6 Sol also shows gains in biology workflows. On GeneBench v1, a benchmark for long-horizon genomics and quantitative-biology analyses, OpenAI said the model outperformed GPT-5.5 while using fewer tokens.

GeneBench v1 (Source: OpenAI)

GPT-5.6 Cybersecurity Capabilities

Cybersecurity is one of GPT-5.6’s biggest advances. OpenAI said GPT-5.6 Sol delivers its strongest cybersecurity performance yet, with improvements in long-horizon tasks such as vulnerability research and software exploitation.

The company said GPT-5.6 Sol achieved performance comparable to Anthropic’s Mythos Preview on ExploitBench while generating only about one-third as many output tokens, a measure of efficiency that can reduce costs and latency.

OpenAI also highlighted results on ExploitGym, a cybersecurity benchmark developed by researchers at the University of California, Berkeley, in collaboration with OpenAI and other frontier AI labs. According to the company, GPT-5.6 Sol, Terra, and Luna all showed stronger cybersecurity capabilities as additional reasoning time was allocated.

Together, the timing puts OpenAI back at the center of a broader fight over who gets access to the most capable AI systems, how those systems are tested before release, and whether governments should have a formal role in reviewing frontier models before they reach the public.

The launch comes as the United States and China race to develop advanced AI models that experts say could accelerate cyberattacks against industries built on older, interconnected systems. Those concerns have pushed Washington to scrutinize major AI model releases more closely, especially systems that may be useful in cyber operations, military planning, or intelligence work.

Chinese authorities have held meetings with top technology companies about potential limits on overseas access to the country’s most advanced AI models, including systems still under development.

OpenAI rival Anthropic recently disabled access to its most advanced AI models, Mythos 5 and Fable 5, after a June 12 U.S. export control order tied to national security concerns. The restrictions were lifted last week after Anthropic added new safeguards. Mythos, which is aimed at cybersecurity professionals, remains available only to some trusted U.S. organizations.

Axios, which first reported the OpenAI launch, said the Trump administration approved a broad rollout of GPT-5.6 after further testing and meetings between OpenAI and government officials.

The White House and the U.S. Department of Commerce did not respond to a Reuters request for comment outside regular business hours.

Before the public rollout, OpenAI limited access to GPT-5.6 to a small group of vetted partners, with their details shared with authorities.

OpenAI previewed the GPT-5.6 models in late June, saying Sol showed stronger agentic performance across coding, biology, and cybersecurity. The company said at the time that GPT-5.6 Sol was competitive with Anthropic’s Mythos Preview on ExploitBench, a cybersecurity benchmark.

The release comes as rivals are moving their own models into the market. Elon Musk said Wednesday that his company, SpaceXAI, is making its leading model, Grok 4.5, available to the public.

For OpenAI, GPT-5.6 is more than a technical upgrade. It arrives at a moment when advanced AI companies are being judged on speed, safety, cost, and their ability to satisfy governments that their systems will not be misused at scale.

U.S. President Donald Trump has signed an executive order creating a voluntary framework for AI developers to provide “covered frontier models” to the U.S. government for up to 30 days before release to trusted partners, Reuters reported.

That framework gives Washington a clearer path to review advanced models before public launch, a shift that could shape how OpenAI, Anthropic, Google, xAI, Meta, and other AI developers handle future releases.

Anthropic has warned that it is “probably impossible” to make any AI model fully secure against jailbreaks. That warning helps explain why model launches are now being treated less like ordinary software updates and more like infrastructure decisions with national security implications.

GPT-5.6 Availability and Pricing

OpenAI said GPT-5.6 will initially be available through its API and Codex for a select group of trusted partners and organizations during the preview period. The company plans to expand access to ChatGPT, Codex, and the API in the coming weeks.

The release also introduces OpenAI’s new model naming system. The number identifies the model generation, while Sol, Terra, and Luna represent capability tiers that can evolve independently. OpenAI said the approach is intended to give developers and businesses clearer choices based on performance, speed, and cost.

Pricing is based on one million tokens. GPT-5.6 Sol costs $5 per million input tokens and $30 per million output tokens. Terra is priced at $2.50 for input and $15 for output, and Luna costs $1 for input and $6 for output.

OpenAI is introducing changes to prompt caching with GPT-5.6, including support for explicit cache breakpoints and a minimum cache lifetime of 30 minutes. Cache writes for GPT-5.6 and newer models are billed at 1.25 times the standard uncached input rate, and cached input reads continue to receive a 90% discount.

GPT-5.6 may still be judged by users on speed, reasoning, coding performance, and price. Yet the bigger story is harder to miss: the most powerful AI models are now moving through a new gatekeeper layer, and that layer increasingly includes the U.S. government.