Appreciate you putting AgentCPM Explore tested in a real local-agent setup. The KV-cache results are especially interesting. Model size is only part of the on-device equation, and memory management can make a big difference for long-context agents. Thanks for the thoughtful deep dive. 🙌
I never heard of OpenBMB before today. A quick look turned into an urgent need to find out more! https://x.com/i/article/2095100288051138560