Reported News
Microsoft Foundry Local runs Qwen3.8-27B as a local coding model inside GitHub Copilot
Microsoft says "Foundry Local brings agentic coding directly to developer hardware": its November 5, 2026 post shows Foundry Local CLI 0.13.0 or later running a 4-bit Qwen3.8-27B (about 18 GB) on NVIDIA GPUs with 24 GB or more of VRAM, connected to GitHub Copilot as a Bring Your Own Key model provider.
- Changes reported
- 6
- listed below, each from a source we read
- We tested
- Not tested yet
- reported from sources; we did not run it ourselves
- Gaps we flag
- 4
- things we could not verify
- Sources
- 1
- 1 primary, linked below
What changed
| # | What the source says | Status |
|---|---|---|
| 01 | Local coding model in Copilot: the post walks through adding Foundry Local as a model provider in the GitHub Copilot app (Bring Your Own Key), pointing it at the local endpoint and picking Qwen3.8-27B in the model picker. | Reported |
| 02 | Needs Foundry Local CLI 0.13.0 or later (installed with winget install Microsoft.FoundryLocal), an NVIDIA GPU with at least 24 GB of VRAM, and CUDA plus cuDNN runtimes installed separately, since Foundry Local does not download them. | Reported |
| 03 | Microsoft says 4-bit quantization brings Qwen3.8-27B to about 18 GB and lists the model's published scores as 73.0 on Terminal Bench 2.1, 61.7 on SWE-bench Pro, 42.3 on NL2Repo-Bench and 42.2 on DeepSWE 1.1. | Reported |
| 04 | The ONNX Runtime pipeline adds DFlash2 speculative decoding, CUDA graph fusions, INT4 weights, KV-cache quantization, CPU offloading, prefix caching and continuous batching for long, concurrent coding requests. | Reported |
| 05 | Against llama.cpp with the same 4-bit model on an RTX 5090 (32 GB) and a DGX Spark (128 GB), Microsoft reports higher generation speed at every tested prompt length, with the gap widening beyond 131K tokens. | Reported |
| 06 | Support for Macs and AMD GPUs, more coding-optimized local models and deeper Copilot integration are described as plans, not shipped features. | Reported |
reported by the sourcemeasured or run by usnot verified
Primary source: Microsoft Foundry blog: Bring Agentic Coding On-Device with Foundry Local. We are the fast, independent index; the vendor’s own page is the source of record for this release.
In the source’s words
“Foundry Local brings agentic coding directly to developer hardware.”
“With 4-bit quantization, the model size can be reduced to approximately 18 GB, allowing it to run well on consumer GPUs with 24 GB or more of VRAM.”
What we have not verified
- We did not install Foundry Local or run Qwen3.8-27B in GitHub Copilot ourselves.
- The llama.cpp speed comparison is Microsoft's own; the post shows charts but we did not reproduce the numbers or check the test settings.
- The benchmark scores are the model's published figures as quoted by Microsoft; we did not re-run them.
- The post shows only the Windows (winget) install path; we did not confirm Linux support for this setup.
This brief is Reported: written from the sources below, not from running the release ourselves.
Sources
- Microsoft Foundry blog: Bring Agentic Coding On-Device with Foundry Localprimary source · fetched 2026-11-05
Related
Spotted an error? See our corrections policy or email the address on the contact page. Markdown version for agents: /news/foundry-local-qwen3-8-27b-github-copilot.md
Want these briefs in your Google Top Stories? Add Agentic Weekly as a preferred source on Google.
Newsletter
The Weekly Verdict. Free, every Monday.
A free email every Monday: the week's few big AI releases, then 5 short use cases for the new models, each with a prompt you can paste.
Free. Double opt-in. One click to unsubscribe.