Qwen 3.5 on M1 Mac: 9B builds a bot, 0.8B runs it
SMRTR summary
A developer successfully ran an agentic coding workflow entirely offline using OpenCode and Qwen3.5 models via LM Studio on a 6-year-old MacBook M1. The setup generated a working Telegram bot that forwards messages to a local AI server, using Qwen3.5-9B for coding tasks and 0.8B for chat responses. While slow, the approach proves viable for sensitive, local development work.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article