Alibaba quietly dropped a big move.

CoPaw-Flash-9B — an AI agent model based on Qwen3.5. With 9 billion parameters, it can run on your own computer.
What's so impressive?
It ties with some benchmarks and Qwen3.5-Plus (closed-source large model).
90 billion parameters vs. hundreds of billions, similar scores.
What excites me even more is the CoPaw framework:
- Supports persistent memory (remembers past conversations)
- Multi-channel connectivity (can connect to Feishu, Discord, etc.)
- Local deployment, no API costs
Qwen3.5's architecture is also very powerful — a total of 397B parameters, with only 17B activated per token. Maximized efficiency.
You don't need an H100 server. A MacBook might be enough to run a personal AI assistant.
I'm thinking of moving some intermediate steps in my content pipeline (data cleaning, format conversion) to the local model, saving over half of the API costs.
By the second half of 2026, personal AI agents could become standard.
But benchmark scores and actual experience are two different things. I'll wait until I get it running locally.