open-source

169Pi releases Alpie Seed, a 9B Apache 2.0 coding model that runs offline on a laptop

Alpie Seed is a 9B-parameter open-weight model of about 5 GB, released Nov 6 under Apache 2.0. It is on Hugging Face, Ollama, LM Studio and the PocketPal app. The company's own SWE-bench Verified score is 61.2%.

Written by Mara at Sig · Nov 6, 2026

169Pi says Seed handles reasoning, code and image or video input, and runs on laptops, recent phones and a Windows PC with no graphics card. Its numbers: 61.2% on SWE-bench Verified against 60% for Qwen3.5-9B, and 72.99% on LiveCodeBench v6 against 65.6% for Qwen3.5-9B. These are from the company's announcement, which says the technical report and benchmark repository are public so you can rerun them.

The company is also opening invite-only access to Alpie Local, a coding agent built on Seed that runs on your machine. It plans a VS Code extension and a desktop agent next. Wider Alpie Local access is promised later this month, with a waitlist at 169pi.ai/alpie-local.

The source for this story is the company's launch coverage, and we haven't found independent test results yet.

Why it matters

A coding model that fits in 5 GB and works with no connection means code never leaves the machine and there is no per-token bill. That only counts if the scores hold up when you run it on your own repo.

Try this

Pull Alpie Seed from Ollama or LM Studio, both listed in the launch, give it one real bug from your repo, and compare the fix with what your current local model produces.