OpenAI Previews GPT-5.6 Sol, Its Most Capable AI Model Yet for Coding, Science and Cybersecurity

OpenAI has unveiled GPT-5.6 Sol, its next-generation frontier AI model, describing it as the company's most capable model to date for coding, scientific research, cybersecurity, and tackling long-running, complex tasks. The company says the new model delivers major improvements in reasoning, instruction following, and agentic capabilities while maintaining a strong focus on safety.

According to OpenAI, GPT-5.6 Sol is designed to handle significantly more sophisticated workflows than previous GPT models. It can sustain longer reasoning chains, execute complex multi-step tasks, and work more effectively across software engineering, scientific discovery, and cybersecurity use cases.

The company says the model has been trained to be more reliable when carrying out extended tasks that require planning, tool use, and iterative problem-solving. OpenAI also claims GPT-5.6 Sol demonstrates stronger performance in coding benchmarks and scientific reasoning, making it suitable for developers, researchers, and enterprise users working on demanding projects.

Designed for Long-Horizon Tasks

One of the key focuses of GPT-5.6 Sol is its ability to complete long-duration objectives that require maintaining context over extended interactions. OpenAI says the model is better at decomposing large problems into manageable steps while adapting its approach as new information becomes available.

The company notes that GPT-5.6 Sol has been optimized for scenarios where users expect an AI system to function more like an autonomous collaborator rather than a simple chatbot. This includes planning, debugging software, analyzing research, and performing complex reasoning across multiple domains.

Improved Coding and Scientific Capabilities

OpenAI says GPT-5.6 Sol delivers substantial gains in software development tasks, including code generation, debugging, repository understanding, and long-form engineering workflows. The model is also intended to assist researchers with scientific reasoning, hypothesis generation, and technical analysis.

The company positions GPT-5.6 Sol as its strongest model yet for cybersecurity-related work, particularly defensive applications such as identifying vulnerabilities, improving software security, and assisting security professionals with analysis and remediation.

Stronger Safety Measures

Alongside capability improvements, OpenAI says GPT-5.6 Sol incorporates its most advanced safety systems to date. The company emphasizes extensive evaluations covering cybersecurity risks, misuse prevention, and harmful capability testing before previewing the model.

According to OpenAI, GPT-5.6 Sol has undergone extensive automated testing alongside expert human evaluations. The company says it has implemented additional safeguards, including refusal training, real-time monitoring, and account review mechanisms to reduce the risk of misuse while preserving legitimate use cases.

OpenAI states that the model is designed to prioritize defensive cybersecurity applications and that its preparedness evaluations did not classify GPT-5.6 Sol as exceeding the company's threshold for dangerous cyber capabilities.

Limited Preview Initially

OpenAI says GPT-5.6 Sol is currently being introduced through a limited preview. During this phase, access is restricted while the company continues testing, gathers feedback, and works through additional deployment processes before expanding availability.

The company says the preview will help validate the model's real-world performance and safety before broader deployment.

OpenAI adds that GPT-5.6 Sol represents another step toward AI systems capable of handling increasingly complex knowledge work while maintaining a strong emphasis on safety, reliability, and responsible deployment.