GitHub Launches Copilot Fusion for Smarter and Cheaper AI Coding

GitHub has launched Copilot Fusion, a system that automatically selects AI models for coding tasks, helping developers balance code quality, speed, and cost.

Instead of choosing a model for every request, users select Fusion and let it decide how to handle the work. It is generally available in Visual Studio Code, Copilot CLI, the GitHub Copilot app, and JetBrains IDEs.

Fusion is a model router and coordinator, not a new standalone model. It checks each request for reasoning, code generation, debugging, and tool needs, then chooses the models and steps needed to produce one final answer. It also considers system health and model availability when making those choices.

Fusion uses three execution patterns. In Single mode, one model completes the task. Cascade starts with an efficient model, then uses a judge to inspect the draft and decide whether a stronger model should take over.

Critique sends a draft to several models for independent reviews before the original model revises its answer using their combined feedback.

These patterns can change within the same conversation. A simple request may use one model, while a difficult follow-up may trigger extra review.

GitHub Launches Copilot Fusion

GitHub says choosing a pattern adds little delay, although Cascade and Critique take longer because they involve additional model calls. Developers can follow progress and inspect which models contributed to the final response.

Fusion Standard selects one model for everyday work. Fusion High supports larger, clearly defined tasks, including difficult bug fixes and changes across several files.

Fusion Max targets work needing deeper reasoning. Fusion Fast serves speed-sensitive requests in Visual Studio Code inline chat and cannot be selected directly by users.

According to GitHub’s documentation, paid Copilot users get a 30% discount on Fusion model costs through November 30, 2026, though High and Max can consume more AI credits than Standard.

Fusion keeps the main conversation on the same model where possible to preserve cached tokens. Supporting models receive only the context they need.

Model selection also follows subscription limits, administrator restrictions, data residency rules, and FedRAMP requirements. Supported local models can participate through Fusion Standard in a public preview.

Extra AI review should not be mistaken for a security guarantee. Earlier reporting on GitHub Copilot prompt injection and the RoguePilot repository takeover flaw shows why untrusted project content remains a concern. Both issues were patched, and neither report establishes a vulnerability in Fusion.

GitHub also warns that discarding a draft does not automatically undo file edits that draft already made. Developers must review workspace changes before committing them. Fusion works separately from subagents and doesn’t launch them, so its role is coordination rather than independent agent management.

Give your SOC the alert context to stops incidents quickly. Integrate Threat Intel from 16K SOCs.

The post GitHub Launches Copilot Fusion for Smarter and Cheaper AI Coding appeared first on Cyber Security News.