OpenAI's GPT-5.6 Is Here: What's New And Why Is It Restricted To US?

OpenAI has announced its next-generation GPT-5.6 family of artificial intelligence models, introducing three new models — GPT-5.6 Sol, Terra and Luna — with improvements in reasoning, coding, biology and cybersecurity. The release, however, will begin with a limited preview for a select group of trusted partners before a wider rollout in the coming weeks.

The launch comes amid increased scrutiny of frontier AI models in the United States, with OpenAI saying it briefed the US government on the models' capabilities ahead of the preview.

Here's all you need to know about the new models.

Three Models, Three Different Use Cases

The GPT-5.6 family introduces a new naming convention. While "5.6" represents the model generation, Sol, Terra and Luna denote capability tiers.

• GPT-5.6 Sol is OpenAI's flagship model, designed for the most demanding reasoning, coding and scientific workloads.
• GPT-5.6 Terra is positioned as the balanced model for everyday enterprise and developer tasks. OpenAI says it delivers performance comparable to GPT-5.5 while costing half as much.
• GPT-5.6 Luna is the company's fastest and most affordable model, aimed at cost-sensitive applications without sacrificing core capabilities.

OpenAI also announced that GPT-5.6 Sol will be available on Cerebras hardware in July, offering inference speeds of up to 750 tokens per second for select customers. OpenAI said Terra offers performance comparable to GPT-5.5 while costing roughly half as much, while Luna is intended for applications where speed and affordability are priorities.

Improvements In Reasoning And Coding

GPT-5.6 introduces two major upgrades designed to tackle increasingly complex tasks.

The first is Max Reasoning Effort, which allows Sol to spend significantly more time reasoning through difficult problems before responding. The second is Ultra Mode, which goes beyond a single AI model by deploying multiple specialised sub-agents that work together on complex workflows, improving performance on coding, research and multi-step tasks.

According to OpenAI, GPT-5.6 Sol recorded improved performance on benchmarks measuring command-line coding workflows, planning and tool coordination. The model also posted stronger performance on GeneBench v1, which evaluates genomics and computational biology tasks while using fewer tokens than GPT-5.5.

The company said GPT-5.6 introduces a new "max reasoning effort" setting that allows Sol to spend more time solving difficult problems before generating a response. It also adds an "ultra mode" that uses multiple sub-agents working together on complex tasks instead of relying on a single model.

Cybersecurity Capabilities See Major Jump

OpenAI highlighted cybersecurity as one of the biggest areas of improvement in GPT-5.6. OpenAI says GPT-5.6 Sol significantly improves vulnerability research and defensive security testing. On ExploitBench, the model reportedly matches Anthropic's Mythos Preview while using only about one-third of the output tokens.

It also performs strongly on ExploitGym, a benchmark created by researchers at the University of California, Berkeley, in collaboration with OpenAI and other frontier AI labs.

However, OpenAI stressed that GPT-5.6 does not cross its "Cyber Critical" threshold under the company's Preparedness Framework. While it can identify software vulnerabilities and exploitation building blocks, it was unable to autonomously generate complete end-to-end exploits during internal testing.

Expanded Safety Measures

OpenAI said GPT-5.6 launches with its most extensive safety framework so far. The company has added multiple layers of safeguards, including:

• Models trained to refuse prohibited cyber assistance.
• Real-time systems that monitor generated responses.
• Additional reviews for higher-risk requests before responses are delivered.
• Account-level monitoring to identify repeated misuse.
• Automated and human "red teaming" to test the models against jailbreak attempts and adversarial prompts.

OpenAI said it spent more than 700,000 A100-equivalent GPU hours on automated red-team testing alongside weeks of human safety evaluations.

Why The Launch Is Limited

Unlike previous releases, GPT-5.6 is currently available only to a select group of trusted partners through the API and Codex.

OpenAI said it briefed the US government on the models before launch and agreed to begin with a limited preview at the government's request. The company described the move as a temporary measure while federal authorities develop a broader framework for evaluating advanced AI models under a recently announced cybersecurity executive order.

OpenAI said it ultimately intends to make GPT-5.6 Sol, Terra and Luna available across ChatGPT, the API and Codex in the coming weeks, adding that it does not believe government preview requirements should become the long-term standard for frontier AI releases.

Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.