OpenAI launches a limited preview of GPT-5.6 for a 'small group of trusted partners'

OpenAI has started previewing its GPT-5.6 series to a limited number of trusted partners, including the US government, ahead of a broader release in the coming weeks. The series includes three variants: Sol, Terra, and Luna, with Sol being the strongest model yet, and Terra offering similar performance to GPT-5.5 at a lower cost. Specifically, Sol is priced at $5 per million input tokens and $30 per million output, while Terra costs $2.50 for input and $15 for output, and Luna costs $1 for input and $6 for output. This move comes after the US government signed an AI cybersecurity order, which asks companies to present their most powerful models for voluntary government review 30 days before making them publicly available.

The GPT-5.6 series is part of a larger trend of AI companies prioritizing safety and security in their models. OpenAI's focus on jailbreak prevention and strengthened protections for high-risk activities reflects the company's efforts to mitigate potential risks associated with its models. For instance, OpenAI spent 700,000 GPU hours to find universal jailbreaks and develop measures against them. This development also highlights the competitive landscape in the AI market, where companies like Anthropic and Google are also working on advanced language models. Anthropic, for example, recently suspended access to its Mythos 5 and Fable 5 models due to concerns about jailbreaking.

The implications of this development are significant, as it suggests that AI companies are taking steps to address potential risks associated with their models. However, it also raises questions about the balance between safety and accessibility, as well as the role of government oversight in the development and deployment of AI models. Specifically, OpenAI's decision to preview its model to a small group of trusted partners, including the US government, raises questions about the potential for government influence over the development and deployment of AI models. Moreover, the fact that Meta was the only holdout from giving the government early access to its AI models, and is reportedly being urged to submit its models for evaluation, suggests that there may be tensions between AI companies and the government over issues of oversight and control.