OpenAI has officially launched GPT-5.6, the company’s newest flagship AI model family, introducing major improvements in reasoning, coding, cybersecurity, scientific research, and autonomous agent workflows. The release marks OpenAI’s biggest model update since GPT-5.5 and signals a broader shift toward AI systems designed to perform complex work rather than simply answer questions.  

Meet the GPT-5.6 Family

GPT-5.6 introduces three models optimized for different workloads:

  • GPT-5.6 Sol — the flagship frontier model designed for maximum capability.

  • GPT-5.6 Terra — a balanced model targeting strong performance with lower cost.

  • GPT-5.6 Luna — the fastest and most cost-efficient model in the family.

OpenAI says this multi-model strategy allows developers and enterprises to choose the right tradeoff between intelligence, latency, and economics for their workloads.  

GPT-5.6 Sol Becomes OpenAI’s New Flagship

The headline model is GPT-5.6 Sol, which OpenAI describes as its most capable model to date.

According to OpenAI, Sol delivers major improvements in:

  • Software engineering

  • Scientific research

  • Long-horizon reasoning

  • Cybersecurity workflows

  • Agentic task execution

  • Tool use and orchestration

The model introduces new reasoning modes including Max and Ultra, allowing the system to spend more time thinking through difficult problems and even coordinate multiple sub-agents for highly complex tasks.  

GPT-5.6 Becomes the Preferred Model for Microsoft 365 Copilot

OpenAI also revealed that GPT-5.6 will become the preferred foundation model for Microsoft 365 Copilot, extending the model’s reach beyond ChatGPT and Codex into Microsoft’s productivity ecosystem. The integration is expected to enhance reasoning, writing, research, and workflow automation capabilities across applications such as Word, Excel, PowerPoint, Outlook, Teams, and Copilot Studio.  

According to OpenAI and Microsoft, GPT-5.6’s improvements in long-context reasoning, instruction following, coding, and autonomous task execution make it particularly well suited for enterprise productivity scenarios. Users can expect more context-aware responses, stronger document generation, deeper analysis, and improved support for multi-step business workflows.  

The move further strengthens the strategic partnership between OpenAI and Microsoft and reinforces Microsoft’s vision of transforming Copilot from an AI assistant into an AI coworker capable of planning, reasoning, and completing work autonomously. As enterprises increasingly adopt agentic AI systems, GPT-5.6 is likely to become one of the core intelligence layers powering the next generation of workplace productivity tools.  

Built for the Agent Era

Perhaps the biggest theme of GPT-5.6 is agentic AI.

OpenAI says the model family was designed specifically for workflows that involve:

  • Multi-step planning

  • Tool usage

  • Long-running execution

  • Parallel task coordination

  • Autonomous software development

This aligns closely with OpenAI’s broader push toward products such as Codex and the newly announced ChatGPT Work platform.  

Better Coding and Security Performance

OpenAI highlighted significant gains in software engineering benchmarks.

GPT-5.6 Sol reportedly sets a new state-of-the-art result on Terminal-Bench 2.1, a benchmark designed to evaluate real-world terminal workflows involving planning, debugging, and tool orchestration. The model also demonstrated stronger biology and cybersecurity capabilities while using fewer tokens than previous generations.  

OpenAI noted that GPT-5.6 Sol achieved competitive results against frontier cybersecurity models while using roughly one-third of the output tokens in certain evaluations, representing a substantial efficiency improvement.  

Safety and Preparedness

With frontier capabilities increasing rapidly, OpenAI says GPT-5.6 ships with its strongest safeguards yet.

The company introduced capability-specific protections and preparedness testing focused on:

  • Cybersecurity

  • Biology

  • Autonomous behavior

  • Tool use

  • Long-horizon planning

According to OpenAI’s preparedness evaluations, GPT-5.6 Sol does not cross the company’s Cyber Critical threshold under current testing conditions, although the company emphasized continued monitoring as capabilities evolve.  

Powering ChatGPT Work

GPT-5.6 also serves as the foundation for ChatGPT Work, OpenAI’s newly announced AI agent platform aimed at knowledge workers and non-technical users.

The platform combines traditional ChatGPT conversations with Codex-style task execution, allowing users to create documents, presentations, websites, spreadsheets, and workflows using natural language instructions.