What Changed — WSL 3 Preview at Build 2026
Microsoft announced WSL 3 on June 2, 2026 at Build. The headline: a paravirtualized hardware access layer replaces WSL 2's full Hyper-V VM, cutting GPU compute overhead from ~15-20% down to 3-5% of bare-metal Linux. More importantly, it exposes the NPU to Linux for the first time—not just the GPU.
The catch: This preview is locked to Copilot+ PCs with Snapdragon X Elite, Intel Meteor Lake, or Lunar Lake NPUs. AMD and discrete NVIDIA desktop setups aren't on the launch list.
What It Means For You — Concrete Impact on Claude Code Daily Use
If you run Claude Code or Aider on a Windows laptop with a local Ollama model, WSL 3 is the upgrade you've been waiting for. Here's the practical difference:
- WSL 2 + NVIDIA desktop: You already have CUDA passthrough. Your
ollama run qwen2.5-coder:14bis using your RTX GPU right now. Stay put. - WSL 3 + Copilot+ laptop: Your NPU and integrated GPU are now accessible from Linux. For local models like Llama 3.2 8B or Qwen2.5-Coder 7B, this means moving from CPU-bound (2-5 tokens/sec) to near-native inference (30-50+ tokens/sec). That's the difference between "unusable" and "daily driver."
For Claude Code specifically: The agent itself calls Anthropic's API, so the GPU isn't doing inference. But if you pair Claude Code with a local model gateway (e.g., for code review, linting, or test generation), or run local tooling that the agent drives, the near-native I/O of WSL 3 cuts friction.
Try It Now — How to Get WSL 3 Working
1. Check if you qualify
# In PowerShell
wsl --version
# If you see WSL version: 2.x.x.x, you're on WSL 2
You need:
- A Copilot+ PC (Snapdragon X Elite, Intel Meteor Lake, or Lunar Lake)
- Windows Insider Program enrollment (Dev or Canary channel)
- The preview build installed
2. Enroll in Windows Insider
Settings → Windows Update → Windows Insider Program → Pick Dev or Canary channel → Install update → Reboot.
3. Install your AI coding stack inside WSL
# Inside WSL (Ubuntu recommended)
# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
# Pull a local model
ollama pull qwen2.5-coder:14b
# Install Claude Code (if you haven't)
npm install -g @anthropic-ai/claude-code
# Install Aider
pip install aider-chat
4. Fix the networking trap
The most common failure: your editor (VS Code with Cline, Continue.dev, or Cursor) can't reach Ollama inside WSL because of the virtual NIC boundary.
Fix it — bind Ollama to all interfaces inside WSL:
# Inside WSL
export OLLAMA_HOST=0.0.0.0:11434
ollama serve
# Find the WSL IP
ip addr show eth0 | grep inet
# Example: 172.20.0.2
Then point your editor at http://172.20.0.2:11434 instead of localhost:11434.
5. Verify GPU/NPU acceleration
# Check if Ollama is using GPU
ollama ps
# Should show model name and GPU utilization
If you see CPU-only, your NPU or GPU isn't being passed through. On WSL 3, this should work automatically on supported hardware.
The Honest Take
If you already run an RTX desktop with WSL 2, your CUDA-backed Aider and Cline setup is fine — stay put. WSL 3 is the real upgrade for Copilot+ laptop owners who want their NPU and GPU available to Linux coding agents without dual-booting. Treat it as preview, not production.
For everyone else: the single biggest bottleneck in your local AI coding workflow on Windows isn't the hypervisor — it's the network boundary between WSL and the host. Fix that first.
Source: dev.to
