Mercury 3
Inception's diffusion language model provides long-context text generation, adjustable reasoning, tool calling and structured outputs through an API.
MAI-Code-1.1-Flash
microsoft.ai
Microsoft's coding model supports agentic development in GitHub Copilot and downloadable local inference on high-memory hardware.
Visit website
Pricing
Not confirmed
Available on
Visual Studio Code, GitHub Copilot CLI, Local inference
Category
Llm Models
What is MAI-Code-1.1-Flash?
MAI-Code-1.1-Flash is Microsoft's coding model for everyday engineering and agentic development. It can plan implementation work, generate code and work from visual references such as screenshots or UI mockups. Microsoft custom-trained it for GitHub Copilot integration in VS Code and Copilot CLI.
Microsoft also offers a downloadable local version with 3-bit quantization and a 256K context window. This route suits developers evaluating on-device coding on high-memory hardware. Experimental local integration in Copilot was announced for the end of October, so availability should be checked before choosing that workflow.
Key features
Agentic coding
Plan and reason through development tasks, progressing from an implementation request toward code changes.
Visual references for code
Microsoft describes generating code from screenshots, diagrams, designs and UI mockups.
Broad language support
Assist with coding across multiple programming languages, frameworks and ecosystems.
Quantized local inference
The downloadable model uses 3-bit quantization while retaining a 256K context window.
Who is it for?
Developers implementing features
Use coding assistance to turn a scoped feature request into proposed changes.
Frontend developers prototyping interfaces
Use a design or screenshot as a reference for an initial implementation.
Engineering teams evaluating local models
Assess whether on-device coding fits available hardware and development workflows.
How to use MAI-Code-1.1-Flash
- Activate a GitHub Copilot plan and install the Copilot extension in VS Code.
- Sign in to GitHub within the editor.
- Update VS Code and the extension; GitHub lists VS Code v1.121 as the model's minimum version.
- Check that your plan and organization policies allow MAI-Code-1.1-Flash, then select it where available.
- Start with a small coding task and review the proposed changes.
MAI-Code-1.1-Flash pricing
Not confirmed
Local model calls have no per-inference charges. Copilot has separate free and paid plans; model access depends on the plan, client and organization settings.
Check current pricing ↗What to know before you use it
- Microsoft recommends more than 120GB RAM for best local performance. This is a recommendation rather than a stated hard minimum.
- Experimental local access in the Copilot app, CLI and VS Code was scheduled for the end of October. Check availability before choosing this integration.
- Copilot model access depends on subscription and organization policies. GitHub lists minimum versions for recent models, and support varies by client.
Frequently asked questions
Is MAI-Code-1.1-Flash available in GitHub Copilot?
GitHub lists it as generally available. Actual access depends on the client, plan and organization settings.
Does local inference make all Copilot usage free?
No. The no-charge statement covers local model calls. Copilot access remains subject to its separate plan terms.
Can I follow the standard Copilot setup to begin?
Yes, for the IDE workflow: use an active Copilot plan, install the extension and sign in to GitHub. Model access remains subject to availability.