← Back to Blog

AI Industry

GitHub Copilot Fusion Nears Opus 5.5 Max at Half the Cost

Oct 29, 2026·3 min read·By Rahish Saifi

Daylit close-up of a developer's hands typing on a silver laptop on a light oak desk beside a coffee mug and a succulent, with the headline COPILOT FUSION, an orange subline Near Opus 5.5 Max at half the cost, and an AIWithRahish badge in the bottom-right corner

At GitHub Universe in San Francisco on Oct. 28, 2026, GitHub made Copilot Fusion generally available—a single Copilot mode that routes and orchestrates underlying models instead of asking developers to pick one model per task. GitHub combines Project HydraFusion's multi-model orchestration with fine-tuned selection from Copilot auto model selection.

What GitHub announced

Fusion collapses model menus into three effort levels:

  • Fusion Standard — versatile models for everyday coding.
  • Fusion High — capable models for challenging tasks and deeper reasoning.
  • Fusion Max — the most capable models for the most demanding tasks and complex reasoning.

Jay Parikh wrote on the Universe blog that Fusion evaluates what a task needs across reasoning, code generation, debugging, and tool use, then chooses how to execute across one or more models. Universe also highlighted Copilot projects for multi-repo agent work, session continuity across clients, and GitHub Enterprise access to Copilot via AI credits—Fusion is the day-to-day routing layer on paid Copilot plans.

How routing and orchestration work

Behind each effort level, GitHub exposes three workflow patterns that describe how models cooperate:

  • Single — one model solves the task directly.
  • Cascade — an efficient model drafts an answer; a quality check accepts it or escalates to a stronger model.
  • Critique — one model drafts, another reviews, and the original model revises.

Cascade and Critique mirror common agent patterns—draft cheaply, verify or revise with a stronger model—without users wiring that graph. Research-preview users of auto, HydraFusion, or HydraFusion Max move automatically to matching Fusion effort levels.

The benchmark claim

GitHub cites the Artificial Analysis Coding Agent Index for external comparison. On that index, GitHub reports that Copilot Fusion Max scored 60—within two points of Claude Code with Opus 5.5 Max—while running at about half the average API cost per task, about half as many tokens, and about 40% less time per task (as GitHub summarizes Artificial Analysis results).

GitHub presents those figures as its summary of Artificial Analysis results—the product goal is strong agent scores without always running frontier models end to end.

Hari Pokhrel, Principal Software Engineer at AT&T, said: "GitHub Copilot Fusion Standard has improved token efficiency across AT&T by automatically selecting the right model for the task, reducing unnecessary spend and preserving capacity for higher-value work."

Admin controls for enterprises

Fusion only uses models enabled in enterprise or organization model policies. Admins can see which models Fusion may pick from, set effective policy per model, and disable models for Fusion without disabling them for everyone else. Enterprise admins can set Fusion at any effort level as the default via Enterprise managed settings.

Billing and availability

Billing stays transparent at the model level: Fusion bills for each model it uses at that model's standard rate. For a limited time, GitHub offers paid Copilot plans a 30% discount on model costs when using Fusion (the JetBrains changelog notes the discount applies for the first 30 days on that client).

Fusion is available on all Copilot plans in VS Code, the GitHub Copilot app, Copilot CLI, and JetBrains IDEs, with more clients coming soon. In JetBrains, Fusion replaces the Auto model option and is rolling out gradually—teams should expect a phased appearance rather than an instant toggle everywhere.

Local-model preview

An experimental opt-in preview lets Fusion choose between supported on-device and cloud models in Copilot CLI, the Copilot app, and VS Code. Managed download is Windows-only; CLI local routing also supports manually configured MAI Code 1.1 Flash or Qwen3.8 Flash Next on a local OpenAI-compatible server on Windows, macOS, and Linux. Routing decisions still happen in the cloud, so sessions are not offline.

What it means for developers

Fusion treats model selection as infrastructure: pick Standard, High, or Max for the job and let Copilot route vendor models underneath—reducing the cost of leaving a frontier model selected all day. For finops and platform teams, automatic routing pairs with admin policy as spend control as agents use more tools and longer sessions.

Teams comparing agent products will note Fusion Max's index score near Opus 5.5 Max; GitHub's pitch is orchestrated efficiency on tokens, time, and dollars. JetBrains users should expect gradual rollout, and the 30% launch discount is a window to benchmark Standard versus Max before enterprise defaults settle.

Sources