Is MAI-Code-1.1-Flash state of the art?Pending benchmarks.Announced Aug 11, 2026; no independent results yet.
Microsoft AI's small-tier agentic coding model, released in GitHub Copilot on Aug 11, 2026 as the successor to MAI-Code-1-Flash, adding image input. On Oct 7 Microsoft announced a quantized build that runs on-device inside GitHub Copilot, with no inference charge for local model calls. A 137B-total / 6.8B-active mixture-of-experts model; no open-weights license stated.
The details so far
- Lab
- Microsoft
- Announced
- Aug 11, 2026
- API
- Not yet
- Open weights
- Unknown
- Parameters
- 137B total / 6.8B active
- Lab
- Microsoft AI (MAI)
- Focus
- Agentic coding in GitHub Copilot; takes screenshots, diagrams and UI mock-ups as input
- Architecture
- Mixture-of-experts, 137B total and 6.8B active parameters
- Context
- 256K tokens
- Predecessor
- MAI-Code-1-Flash (launched June 2026 at Microsoft Build); GitHub retired it from Copilot on Sep 10, 2026
- Availability
- GitHub Copilot since Aug 11, 2026: Copilot CLI, cloud agent, Copilot app, Copilot Chat on GitHub, VS Code, Visual Studio, GitHub Mobile, JetBrains, Eclipse and Xcode. Free and Student plans get it through auto model selection; Business and Enterprise admins must enable its policy
- Price
- 73% lower list price than MAI-Code-1-Flash; 0.25× premium request multiplier for annual Copilot subscribers
- On-device
- Announced Oct 7, 2026: a build quantized to about 3.3 bits per weight (53 GB, peak 75.5 GB memory at 256K context). Microsoft recommends more than 120 GB of RAM. Experimental access in the Copilot app, Copilot CLI and VS Code by the end of October, via Copilot's Auto routing, the Windows ML provider or an OpenAI-compatible local endpoint. No inference charge for local model calls
- Weights
- The local build is downloadable to run on device; no open-weights license stated
- API
- No standalone API stated; used through GitHub Copilot
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: higher-quality code than MAI-Code-1-Flash at a quarter of the price, using 25% fewer tokens per task with tokens streaming 25% faster in GitHub Copilot.
Lab claim: 22% better on Terminal-Bench 2.1 in GitHub Copilot CLI and 15% better on .NET tasks than MAI-Code-1-Flash; code survival up 4% and return visits up 9% in production.
Lab claim (self-reported): the quantized on-device build scores 70.8% on SWE-Bench Verified and 66.29% on Terminal-Bench 2.1, against 72.6% and 62.9% for the BF16 cloud model.
Sources
LABMicrosoft AI: MAI-Code-1.1-Flash: Better, faster, at a quarter of the cost (updated Oct 7)microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/ · Aug 11, 2026LABGitHub Changelog: MAI-Code-1.1-Flash available in GitHub Copilotgithub.blog/changelog/2026-08-11-mai-code-1-1-flash-available-in-github-copilot/ · Aug 11, 2026LABMicrosoft AI on X: on-device model calls inside GitHub Copilotx.com/MicrosoftAI/status/2107926901150908483 · Oct 7, 2026LABBringing local models and sandboxed tools to Windows and GitHub Copilotcommandline.microsoft.com/local-models-sandboxed-tools-github-windows/ · Oct 7, 2026LABMAI-Code-1.1-Flash model page (Microsoft AI)microsoft.ai/models/mai-code-1-flash/LABSupported AI models in GitHub Copilot (GitHub Docs)docs.github.com/copilot/reference/ai-models/supported-modelsTracking ends Jan 20, 2027 unless the API or weights become available.