4gpts retweeted
Introducing Qwen3.8-27B-pi-adapted 🤗
We adapted Qwen's model specifically for the pi coding agent using reinforcement learning and a LoRA adapter.
The results, measured as pass@1:
🚀 SWE-bench Pro (Hard 51): 29.4% → 62.7%
💻 Terminal-Bench 2.1: 39.5% → 55.8%
⚡ 55–62% fewer tool calls per task
This is the power of open models: you can adapt them to YOUR tools and YOUR use cases. Let’s build! 💛
Open weights: huggingface.co/harness-hub/Q…
Methodology: huggingface.co/blog/harness-…
Shopping gets that surface first. 86% of shoppers who used AI to research a purchase checked its recommendation somewhere else (Product.ai, 1,463 US shoppers, 2026). A stated spec turns that check into a comparison a surface can show.
1
2
A personal AI is personal only if you own it.
Meta's Muse shops for you and pays with a one-time card. It works. And it lives on a computer Meta runs, under a policy that, in Meta's own words, "does not prevent Meta from accessing data" to support, secure or operate the service.
1
21
Where each part of Muse lives, and the record in full:
4gpts.com/en/resources/meta-…
1
17


