MAI-Code-1-Flash: Microsoft's First In-House Copilot Coding Model
GitHub CopilotAt Microsoft Build 2026, GitHub launched MAI-Code-1-Flash inside GitHub Copilot, the first coding model Microsoft has built entirely in-house, independent of OpenAI. The 5-billion-parameter model outperforms Claude Haiku 4.5 across all four core coding benchmarks, including a 16-point lead on SWE-Bench Pro (51.2% vs. 35.2%), while using up to 60% fewer tokens on complex tasks. Available at no additional cost to all plan tiers (Free, Pro, Pro+, and Max), the rollout begins with a limited set of VS Code users and expands gradually over the coming weeks.
Featured Video
A video we selected to help illustrate this changelog
Sources & Mentions
5 external resources covering this update
Microsoft Build 2026 Day 2 LIVE | GitHub Copilot, VS Code, and more
Youtube
MAI-Code-1-Flash
Hacker News
Biggest Microsoft Build 2026 announcements
Tom's Guide
Microsoft Build 2026: MAI-Thinking-1 Is First In-House Reasoning Model, Trained Without OpenAI Data
Link
Microsoft MAI-Thinking-1 & MAI-Code-1-Flash: Developer Guide to 7 New MAI Models
Dev.to
MAI-Code-1-Flash: Microsoft Builds Its Own Copilot Coding Model
GitHub announced MAI-Code-1-Flash on June 2, 2026, at Microsoft Build, marking a significant strategic inflection point: Microsoft is no longer exclusively relying on OpenAI models to power GitHub Copilot. MAI-Code-1-Flash is the first in a new wave of purpose-built coding models developed entirely by Microsoft, trained on production Copilot harnesses and licensed data.
What Is MAI-Code-1-Flash?
MAI-Code-1-Flash is a 5-billion-parameter coding model designed for fast, efficient assistance on everyday developer workflows. Despite its compact size, the model incorporates adaptive reasoning, staying concise for simple requests while spending more reasoning budget on complex tasks. This allows it to punch above its weight class in agentic and multi-file scenarios.
The model was purpose-built for the GitHub Copilot environment rather than adapted from a general-purpose foundation model. Training on production Copilot usage patterns gives it a native understanding of how developers actually interact with AI coding assistants.
Performance
The model outperforms Claude Haiku 4.5 on all four core coding benchmarks tested at launch:
- SWE-Bench Pro: 51.2% vs. 35.2% (16-point lead)
- Up to 60% fewer tokens on harder SWE-Bench Verified tasks
The token efficiency gain is particularly significant in the context of usage-based billing: developers get the same (or better) results for meaningfully lower token consumption, which translates directly to lower costs per agentic session.
Strategic Context
MAI-Code-1-Flash arrives months after Microsoft and OpenAI concluded their seven-year exclusive partnership. The timing underscores Microsoft's intent to build model independence into its developer tooling stack. The company described the model as "the first in a new wave," signaling that more Microsoft-built models are in development for the Copilot ecosystem.
Availability
The rollout begins with a limited set of users on Copilot Free, Pro, Pro+, and Max plans, accessible via the model picker in Visual Studio Code. Broader IDE support and expanded availability are expected in the coming weeks. The model carries no additional cost on existing Copilot plans.