At GitHub Universe 2026 on Oct 28, GitHub made Copilot Fusion generally available. It merges the multi-model orchestration of the Project HydraFusion research preview with Copilot auto model selection: users pick Copilot Fusion and an effort level (Standard, High or Max) in the model picker, and the system decides which models to use and whether to run a Single, Cascade or Critique workflow. GitHub says Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, two points below Claude Code with Opus 5.5 Max, at about half the API cost per task, about half the tokens and about 40% less time. Each underlying model is billed at its standard rate, and paid plans get 30% off through Nov 30.
Key Takeaways
- ✓Scores (Artificial Analysis Coding Agent Index v2.0): Fusion Max 60 versus 62 for Claude Code with Opus 5.5 Max; $5.59 per task versus $11.42, while the Devin Fusion pairing scores 62 at $8.00
- ✓Breakdown: a slight lead on Terminal-Bench 4.0 at 71, third on FrontierCode at 51 (Claude Code 56, Devin Fusion 54), and a weaker 57 on SWE-Atlas-QnA-AA
- ✓Efficiency: about 7.2M tokens per task, roughly half of Claude Code with Opus 5.5 and 30% fewer than Devin Fusion; GitHub reports about 40% less time per task
- ✓The pool includes Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra and GPT-6 Sol; admins can disable individual models for Fusion only and set Fusion as the default
- ✓Availability: all Copilot plans in VS Code, the Copilot app, Copilot CLI and JetBrains IDEs; existing auto, HydraFusion and HydraFusion Max selections migrate to matching effort levels
Key Decision Metrics at a Glance
Heavy Claude Code use: compare subscription limits and API bills
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and pain point
Coding agents keep getting stronger, but picking a model has become a daily chore: developers reach for the most expensive model on a hard task and forget to switch back, and spend climbs. GitHub put HydraFusion and HydraFusion Max into research preview in mid-October; on day one of Universe 2026 it merged them with auto model selection into Copilot Fusion and made it generally available. An AT&T engineer quoted in the changelog says Fusion Standard cut unnecessary token spend.
Architecture and mechanism
- Fusion is a runtime router and orchestrator that picks an execution pattern per prompt: Single, where one model answers directly; Cascade, where an efficient model drafts and a judge accepts it or escalates to a stronger model; and Critique, where one model drafts, several models critique in parallel, and the draft is revised.
- Model switches happen at cache boundaries to avoid extra cache costs, and simple tasks skip the review calls.
- Fusion has no context window of its own; the displayed limit is the smallest among the models it uses.
Benchmarks
On the Artificial Analysis Coding Agent Index v2.0 (third-party, Max setting in Copilot CLI), Fusion scores 60 versus 62 for Claude Code with Opus 5.5 Max, at $5.59 versus $11.42 per task. It slightly leads Terminal-Bench 4.0 at 71, ranks third on FrontierCode at 51, and lags on SWE-Atlas-QnA-AA at 57. The "40% faster, up to 77% cheaper" figures are GitHub's own claims.
Getting started
Update VS Code, the Copilot app, Copilot CLI or the JetBrains plugin, choose Copilot Fusion in the model picker, then pick Standard, High or Max. Use Standard for everyday edits, High for multi-file fixes and Max only for the hardest problems. Each model is billed at its standard rate, with 30% off on paid plans through Nov 30. The same day GitHub also shipped Copilot projects and Copilot code review Max, and said GitHub Enterprise now includes Copilot without a separate subscription.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.