GitHub Makes Model Routing the Default Choice in Copilot

Copilot Fusion replaces the model picker with an effort level

Instead of choosing a model per task, a user selects Copilot Fusion and picks an effort level: Standard, High or Max. Copilot then chooses which models handle the work. GitHub says Fusion combines its HydraFusion research preview's multi-model orchestration with Copilot's existing auto model selection. For each prompt it uses one of three workflows. In Single, one model solves the task. In Cascade, "an efficient model drafts a response", and a quality check accepts it or escalates to a stronger model. In Critique, one model drafts, another reviews and the first revises.

Fusion is available on all Copilot plans in VS Code, the Copilot app, Copilot CLI and JetBrains IDEs. Billing is per model used, at each model's standard rate, with a limited-time 30% discount for paid plans. GitHub reports that Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, "within two points of Claude Code with Opus 5.5 Max", at "about half the average API cost per task", about half the tokens and about 40% less time per task. That is GitHub's own report of the result.

Administrators set the menu

Fusion "only uses the models that are enabled in your enterprise and organization model policies". Enterprise administrators can see which models it picks from, set a policy for each model's use within Fusion, and make Fusion the default through managed settings. In practice, the person who chooses models becomes the administrator who writes the policy, not the developer at the prompt.

Agentic Workflows reach general availability

GitHub Agentic Workflows, an open source project from GitHub Next, runs coding agents inside GitHub Actions for tasks such as issue triage and CI failure analysis. Teams write the automation in Markdown, and it compiles to standard Actions YAML. The security model is layered. Agents read GitHub content under integrity filter rules, "run with read-only permissions by default", and execute in a sandboxed container behind the Agent Workflow Firewall. Their outputs go through a safe outputs check, and "a dedicated threat detection job scans all proposed changes before they are applied". General availability adds self-hosted ARC runners, a non-experimental gh aw forecast for spend ranges, model routing by task type, and optional micro-VM isolation. Using the default Copilot CLI engine requires a Copilot license.

What remains uncertain. The Fusion benchmark comes from GitHub, and how often Cascade escalates on real repositories is not reported. Because billing is per model used, a task that escalates costs more than the draft alone, so the bill is harder to predict. The discount's end date is not given.

Analysis: Routing moves two decisions out of the developer's hands: which model runs, and what it costs. Both become policy. A team at Belgavi's Book Store adopting Fusion should first decide which models are allowed at all, then compare a week of Fusion spend against fixed-model spend on its own pull requests. For Agentic Workflows, the useful pattern is that the agent proposes and a separate job checks before anything changes. A triage agent that labels CASE-1042 as a refund issue should write a proposed label, not apply one, until the check passes.

Sources and evidence

• Route and orchestrate models with GitHub Copilot Fusion (GitHub Changelog, October 28)
• GitHub Agentic Workflows generally available (GitHub Changelog, October 28)
• Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index, with Coding Agent Index results (Artificial Analysis, October 28)