OpenAI unveils GPT-5.6 AI models with phased rollout and stronger safety checks

OpenAI has introduced GPT-5.6, its latest generation of artificial intelligence (AI) models. Instead of launching a single model, the company has released a family of three models, Sol, Terra and Luna, designed for different levels of performance, speed and cost.

However, unlike previous launches, GPT-5.6 is not yet available to all users. The company is rolling it out first to a small group of trusted partners before making it available more broadly through ChatGPT, the API and Codex.



OpenAI’s new AI models and their capabilities:

The flagship model, GPT-5.6 Sol, is designed for tasks that require deeper reasoning and problem-solving. OpenAI says it performs better than its earlier models across areas such as coding, biology and cybersecurity. The company has also introduced a new "Max" mode, which allows the model to spend more time thinking before generating an answer.

Another addition is "Ultra" mode, which uses multiple AI agents working together to solve more complex problems that would be difficult for a single model to handle.

Alongside Sol, OpenAI has launched GPT-5.6 Terra and GPT-5.6 Luna.

Terra is aimed at users who need strong performance at a lower cost, with OpenAI claiming it offers capabilities close to GPT-5.5 while costing roughly half as much. On the other hand, Luna is the smallest and fastest model in the family and is designed for applications that need quick responses and can process large numbers of requests efficiently.


Why OpenAI is limiting ChatGPT-5.6 access:

OpenAI says GPT-5.6 comes with stronger safety checks. The models are designed to reject harmful cyber requests, while monitoring systems can review, delay or block risky responses. OpenAI says the safeguards are meant to help researchers find software flaws without enabling misuse.

One of the biggest differences with this release is how OpenAI is making it available. The company says it limited the initial rollout after discussions with the US government about the safe deployment of advanced AI systems. As part of the preview, access has been restricted to a small number of organisations whose participation has been shared with government officials. OpenAI has said it does not expect this process to become the standard for future launches, but agreed to the temporary arrangement while broader policies around frontier AI are developed.

Despite its improved capabilities, OpenAI says GPT-5.6 still has several limitations. According to the company, the internal testing showed that the model is better at identifying software vulnerabilities than carrying out complete cyberattacks on its own. The additional safety systems may also make the model refuse more requests or respond more slowly than earlier versions.

The release of GPT-5.6 reflects a major transformation in the AI industry. Companies are no longer focused only on making models smarter and more capable. They are also placing greater emphasis on controlling how these systems are deployed and reducing the risks of misuse. OpenAI’s phased rollout and added safeguards suggest a more cautious approach to releasing its most advanced models.