OpenAI Unveils GPT-5.6 Lineup: From Flagship Sol to Economical Luna

OpenAI has announced a limited preview of its new next-generation model lineup: GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna. These models are designed to provide maximum flexibility and broad accessibility, covering both flagship and economical use cases.

GPT-5.6 Sol is positioned as a flagship product that, according to OpenAI, surpasses the previous GPT-5.5 model by an order of magnitude. This model is designed for the most complex and demanding tasks, setting a new standard of excellence, especially in the fields of cybersecurity and command-line operations. Specifically, GPT-5.6 Sol demonstrates outstanding results on the specialized benchmark Terminal-Bench 2.1, which evaluates the model's ability to perform complex command-line workflows requiring not only planning but also iteration, and the coordination of various tools. This means that Sol can handle multi-stage, complex tasks, simulating the work of a highly skilled specialist who must sequentially solve problems using various digital tools.

While Sol is the peak of performance, OpenAI also introduced two more specialized and accessible models. GPT-5.6 Terra is developed as a balanced option. It provides performance that, according to the company, is comparable to GPT-5.5, but significantly reduces the cost of use—by half. This makes it an ideal choice for routine yet demanding business work where a balance between result quality and cost-effectiveness is important. GPT-5.6 Luna, conversely, is the most economical model in the lineup. It is aimed at tasks requiring large volumes of data processing, where low cost and high speed are critical, even if absolute maximum performance, like that of Sol, is not required. Thus, OpenAI solves the 'one size fits all' problem, offering three clearly differentiated tools for different budgets and levels of task complexity.

Since OpenAI believes in the broad accessibility of its developments, they plan to make all three models—Sol, Terra, and Luna—publicly available in the coming weeks. However, access is currently limited to a preview provided upon request by the US government and a limited circle of trusted partners through Codex and API. This indicates that, despite the stated goal of broad accessibility, the initial launch will be controlled and targeted at key corporate and government clients.

Special attention is paid to security. GPT-5.6 Sol launches with the most robust security system to date. OpenAI has significantly enhanced real-time protection to counter high-risk cyber-activity and prevent model misuse. To achieve this level of protection, the company spent weeks 'hardening' the system using red teaming, which is a standard but critically important practice in the development of advanced AI systems. This ensures that even the most powerful capabilities of Sol are used within defined ethical and legal boundaries, minimizing risks associated with potential misuse in critical areas such as cybersecurity. The flagship Sol model, being the most powerful for cybersecurity tasks, requires a corresponding level of control and testing, which has been implemented.