Copilot Rivals Claude Code While Offering 11 Claude Models

GitHub said on 28 October 2026 that, in the Artificial Analysis Coding Agent Index, “Copilot Fusion Max scored 60, within two points of Claude Code with Opus 5.5 Max”. Anthropic’s Claude Code is the agent GitHub chose to measure Copilot against, and Anthropic is also the largest supplier of models inside Copilot: 11 of the 28 models GitHub lists as generally available.

GitHub said Fusion delivered that score “at about half the average API cost per task, using about half as many tokens and taking about 40% less time per task”.

Business Pill · The Model and the Harness

An evergreen Business Engineer explainer on the model and the harness built around it. It explains the general idea only and does not discuss GitHub Copilot.

What Copilot Fusion Is

GitHub’s documentation describes Copilot Fusion as “an intelligent runtime model router and orchestrator”. It says: “You select Copilot Fusion in the model picker, but you aren’t selecting a specific model. You’re choosing a system that routes and orchestrates models at runtime.”

GitHub lists three execution patterns: a single model; a cascade, where “An efficient model drafts a solution, and a judge inspects the draft and decides whether to accept it or escalate to a stronger model”; and a critique pattern, where several models review a draft in parallel.

GitHub does not publish a fixed list of the models Fusion uses. Its documentation says Fusion “chooses from the models available, subject to your policies and subscription type” and that “The model pool changes as new models become available and as GitHub evaluates which models perform best.” The two models it names are local ones: Fusion can use “MAI Code 1.1 Flash or Qwen3.8 Flash Next” where installed.

What Was Posted on X

TestingCatalog’s post lists five GitHub Universe ’26 announcements: Copilot Fusion, Copilot Projects, a security extension for GitHub Copilot, Copilot dreaming and Copilot Automations.

The 28 Models Inside Copilot

GitHub’s table of supported models lists 28 as generally available: 11 from Anthropic, 10 from OpenAI, three from xAI, two from Google, one from Microsoft and one from Moonshot AI. The Anthropic list includes Claude Opus 5.5, the model Claude Code used in GitHub’s comparison. The Google entries are Gemini 3.8 Flash and Gemini 4 Argon; Microsoft’s is MAI-Code-1.1-Flash; Moonshot’s is Kimi K3.

Microsoft chief executive Satya Nadella said in his official prepared remarks that “MAI Code Flash had 3 million users in GitHub Copilot last month alone, helping developers get more out of their AI credits.”

What Microsoft Said on the Call

Nadella said in his prepared remarks that Microsoft announced “broad availability of Copilot Fusion, which is all about bringing multi-model orchestration to software development”, and that “It delivers frontier-level quality up to 40 percent faster, and at up to 77 percent lower cost.”

Answering an analyst, Nadella gave slightly different figures, according to a machine transcript of the webcast. He said Artificial Analysis “had what I think is the real benchmark, which is the coding agents using our Fusion, and I think we were number two and number three there”, and “You can now get frontier performance 45% faster, 75% cheaper.”

He added, per the same transcript: “If you’re a coding agent user, you want, in fact, Anthropic and open AI, by the way, to compete for your workload, not once a quarter, but for every task.”

Where Microsoft Says Copilot Is Growing

Nadella said “all major coding agents have showed up to work on GitHub”, that “Agent generated PRs have increased 6X over the past six months” and that “agents now open more than a third of all PRs on the platform.” He said “Credit consumption from agentic features on GitHub has more than doubled since usage-based billing went into effect” and that “GitHub Advanced Security revenue increased over 50 percent year over year”.

Chief financial officer Amy Hood said in her prepared remarks that Microsoft 365 commercial cloud revenue growth was 17% on a reported basis, and that “Results were ahead of expectations driven by stronger GitHub Copilot adoption and usage”.

Describing revenue per user, Hood said growth was led by premium seat-based offerings, “with additional contribution from usage-based offerings, primarily GitHub Copilot”.

Our Arithmetic

By our arithmetic, Anthropic supplies 39% of the 28 models GitHub lists as generally available in Copilot (11 of 28), OpenAI 36% (10 of 28) and Microsoft 4% (one of 28).

By our arithmetic, if Claude Code with Opus 5.5 Max scored the full two points higher, Fusion Max’s 60 would be about 97% of its score, reached at what GitHub says is about half the API cost per task.

Connecting the Coverage

We covered the launch in GitHub’s Copilot Fusion nears Opus 5.5 at half the cost. The pricing shift behind it is in Microsoft takes Copilot past per-seat, and Microsoft’s own models are in Microsoft aims to match leading AI in 3-6 months.

The Structural Read

As we read GitHub’s documentation, Fusion is not a model but a router that picks from the models a customer’s plan and policies allow, so its rivals’ models are part of what it sells.

As we read the call, Microsoft framed the routing itself as the product: Nadella said, per the machine transcript, that coding-agent users want Anthropic and OpenAI to compete for each task.

Microsoft’s own model in the catalogue is MAI-Code-1.1-Flash; Nadella said MAI Code Flash had 3 million users in GitHub Copilot last month.

Three Implications

FOR ANTHROPIC By our arithmetic, Anthropic supplies 39% of the 28 models GitHub lists as generally available in Copilot, while GitHub benchmarks Copilot against Claude Code.

FOR PRICING GitHub says Fusion Max reached its score at about half the average API cost per task, and bills each model Fusion uses at that model’s standard rate.

FOR WHAT IS NOT DISCLOSED GitHub does not say which models produced Fusion’s index score or how many developers use Fusion.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Agentic Harness War.

The framework argues: “Foundation models become relatively commoditized as orchestration layers capture differentiation.”

Copilot Fusion is GitHub’s orchestration layer; GitHub says it routes among the models available to a customer from Copilot’s catalogue, which GitHub lists as 28 generally available models from six providers. The framework was published in July 2026 and does not discuss Copilot Fusion.

What Is Not Established

GitHub does not say which models Copilot Fusion used to reach its index score, and it publishes no fixed Fusion model pool. The Artificial Analysis figures here are as GitHub reported them; we could not read the scores from Artificial Analysis’s own page, which renders its charts in the browser. Nadella’s ranking claim and his 45% and 75% figures come from a machine transcript of the webcast; his official prepared remarks say up to 40% faster and up to 77% lower cost. Neither company has said how many developers have moved from Claude Code to Copilot, or the reverse.

The Bottom Line

GitHub measures Copilot Fusion against Anthropic’s Claude Code, and says Fusion Max scored 60, within two points, at about half the API cost per task. Anthropic also supplies 11 of the 28 models GitHub lists in Copilot, including the Opus 5.5 model used in that comparison.