Breaking

Microsoft adds local MAI-Code-1.1-Flash calls to GitHub Copilot with no inference charge

October 7, 2026 at 16:29 EDT

Software Dev & Coding
AI Agents
Foundation Models

Microsoft is bringing MAI-Code-1.1-Flash onto developer devices through GitHub Copilot, giving Copilot's router an on-device coding model option with no inference charge for local model calls. The October 7 update positions the model as a hybrid coding assistant: eligible work can run on local hardware, while harder tasks can still be routed to cloud models when more capability is needed.

October 2026 · GitHub Copilot

MAI-Code-1.1-Flash goes local

Copilot is getting an on-device coding model path: local model calls can run on supported hardware with no per-inference charge, while cloud models remain available for harder work.

256K

local context window

>120GB

RAM recommended

25%

greater token efficiency

22%

Terminal-Bench 2.1 gain

Cost shift: from cloud meter to local call

MAI-Code-1-Flash

MAI-Code-1.1-Flash

Version 1.1 is described as one quarter of the cost of the June model, with local calls adding a zero-inference-charge path for eligible work.

How the hybrid path works

1. Route task

Copilot evaluates the coding request.

→

2. Run local

Eligible calls use device hardware.

→

3. Escalate cloud

Harder tasks can still use cloud models.

What improves

Lower repeated-call cost, local processing for supported workflows, and faster iterative coding loops.

What limits adoption

Best performance requires more than 120GB of RAM, and experimental routing must prove reliable.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.