GitHub Copilot Fusion Is Here, Combining AI Models to Cut Coding Costs by 50%

GitHub has introduced Copilot Fusion at GitHub Universe 2026, a new AI coding feature that automatically selects and combines models from different providers to improve performance while reducing costs.

According to GitHub’s official announcement, Copilot Fusion builds on the platform’s automatic model selection technology and Project HydraFusion’s multi-model orchestration capabilities.

The system can use AI models from OpenAI, Anthropic, Google, and other providers, choosing the best approach based on a coding task’s complexity.

GitHub Copilot Fusion offers three reasoning levels

Rather than selecting individual AI models, developers can choose Copilot Fusion directly from the model picker.

GitHub offers three options: Fusion Standard for everyday coding tasks at lower costs, Fusion High for more complex tasks that need extra reasoning, and Fusion Max for demanding tasks that require the most powerful AI models.

Copilot Fusion automatically determines whether a task needs a single model or collaboration between multiple models.

To accomplish this, GitHub uses three execution patterns. The Single approach assigns the entire task to one model, while Cascade starts with a smaller model and evaluates whether a more capable model should take over.

The third approach, Critique, allows one model to generate a response while other models review its output and suggest improvements before the system produces a revised answer.

Copilot Fusion Max challenges Claude Code at half the cost

GitHub also shared benchmark results showing that Copilot Fusion Max delivers competitive coding performance while using fewer resources.

On the Artificial Analysis Coding Agent Index, Fusion Max scored 60 points, just two points behind Claude Code running Opus 5.5 Max.

Despite the small performance difference, GitHub reported that Fusion Max achieved approximately 50% lower average API costs per task, consumed 50% fewer tokens, and completed tasks around 40% faster.

GitHub brings local AI models into Copilot Fusion

GitHub is also experimenting with routing coding tasks between cloud-based AI models and models running directly on developers’ computers.

On supported Windows devices, Copilot Fusion Standard can automatically use Microsoft’s MAI Code 1.1 Flash model locally.

This approach could reduce cloud processing requirements for supported tasks while allowing developers to access more powerful cloud models when necessary.

Enterprise administrators can also control which models Fusion can access and configure it as the default AI coding experience across their organizations.

Copilot Fusion is available now with a 30% discount

GitHub Copilot Fusion is generally available across Visual Studio Code, GitHub Copilot CLI, the GitHub Copilot app, and JetBrains IDEs.

GitHub charges developers for the individual AI models used during each task, without adding a separate orchestration fee.

Microsoft is also offering a 30% discount on model costs through November 30, 2026, giving developers an opportunity to try the new orchestration system at a lower price.

The launch comes as Microsoft continues expanding its AI model portfolio. The company recently made its MAI-Image-2.6 and MAI-Image-2.6-Flash models generally available in Microsoft Foundry, while its newer MAI-Image-2.7 model climbed AI image-editing rankings.

Copilot Fusion represents another step toward AI development tools that select and coordinate models automatically, rather than requiring developers to choose a single model for every task.