GPT-5.6 becomes the preferred model in Microsoft 365 Copilot

OpenAI has publicly launched the GPT-5.6 family of models following clearance from the US government. The release introduces three distinct tiers: Sol, Terra and Luna. This tiered approach allows users to match the model size and cost to specific professional workflows. The launch arrives at a pivotal moment in the AI industry, as multiple frontier models have been released within a single week.

The release week has seen significant activity in the frontier model space. GPT-5.6 Sol is positioned as the flagship model for complex reasoning and long-horizon tasks. Terra serves as a balanced model for everyday work, offering strong performance at a moderate price point. Luna provides a highly cost-efficient option for scaled deployments where cost per token is the primary constraint.

GPT-5.6 Sol has established a new standard for coding capabilities. On the Artificial Analysis Coding Agent Index, Sol with maximum reasoning settings achieved a score of 80. This result places it 2.8 points above Claude Fable 5. The model achieved this whilst using less than half the output tokens and taking less than half the time of its predecessor.

The efficiency gains extend across the model family. OpenAI reports that Sol is 54 per cent more token efficient on coding tasks compared to previous models. This improvement has direct implications for the cost of running agentic systems. When a model can solve the same problem using half the tokens, the cost per task is cut in half.

On the Agents' Last Exam evaluation, which tests long-horizon professional workflows across 55 different fields, Sol scored 53.6, eclipsing Fable 5 by 13.1 points. Even at medium reasoning settings, Sol beats Fable 5 by 11.4 points at roughly one-quarter the estimated cost.

The three-tier structure reflects a deliberate strategy to serve different use cases. Sol is designed for frontier reasoning and long-horizon agentic work. Terra is positioned as a balanced model that offers GPT-5.5-competitive performance at twice the token efficiency. Luna is optimised for cost-sensitive applications where performance can be traded for affordability.

The tiered structure reflects a simple reality: not every task requires frontier intelligence. Many operational workflows can be handled effectively by smaller models at a fraction of the cost. By offering multiple tiers, OpenAI enables teams to optimise their AI expenditure based on task complexity.

In a significant enterprise development, GPT-5.6 is now the preferred model in Microsoft 365 Copilot. The model powers features across Word, Excel, PowerPoint, Chat and Cowork. This integration brings frontier intelligence directly into the applications that project delivery professionals use daily.

The partnership aims to help knowledge workers manage complex, multi-step tasks.

For AEC firms, this integration carries particular weight. Most project delivery teams already use Microsoft 365 extensively. The ability to access frontier AI capabilities directly within these applications eliminates the need to switch between tools or copy content between systems.

Industry leaders have noted the practical improvements in the new model family. The focus on persistence and efficiency appears to be yielding tangible results in real-world applications.

The introduction of Programmatic Tool Calling in the Responses API allows the model to filter large amounts of intermediate data. It retains only the relevant information and adapts its workflow automatically. This feature reduces prompt tokens by up to 38 per cent with no loss in quality.

For complex project workflows, this efficiency gain is significant. When a model can filter irrelevant data automatically, fewer tokens are consumed and the task completes faster. This compounds over time when running multiple agentic tasks.

OpenAI describes GPT-5.6 as its strongest cybersecurity model to date. The system layers protections trained into the model with real-time checks and monitoring. Access is calibrated to trust and risk levels.

For project delivery professionals handling sensitive commercial data and intellectual property, the security focus is particularly relevant. The model supports defensive cybersecurity activities, including threat modelling, code review and patching.

We recognise that for AEC firms, the combination of Microsoft 365 integration and improved token efficiency presents a compelling case for adoption. The ability to execute long-running tasks with fewer model round trips directly impacts the cost of running agentic workflows.

The tiered pricing structure also enables firms to optimise their approach. Complex strategic tasks can use Sol. Routine operational tasks can use Luna. This flexibility allows teams to manage their AI expenditure more effectively.

• The tiered pricing structure of Sol, Terra and Luna allows firms to optimise their AI expenditure based on task complexity, matching model capability to specific use cases.

• Integration as the preferred model in Microsoft 365 Copilot embeds advanced reasoning directly into standard enterprise workflows, eliminating the need for tool-switching.

• Significant improvements in token efficiency reduce the overall cost of executing long-horizon agentic tasks, with Sol achieving 54 per cent better efficiency than predecessors.

• Enhanced cybersecurity capabilities provide greater assurance for firms processing sensitive project documentation and commercial data.

• Programmatic Tool Calling reduces prompt tokens by up to 38 per cent, allowing models to filter irrelevant information automatically and complete tasks faster.

Subscribe to the Project Flux newsletter for deeper analysis on frontier model releases and their implications for your practice.

All content reflects our personal views and is not intended as professional advice or to represent any organisation.