Running local LLMs is not going to replace your Claude or Codex subscription.. for now.
You are running local LLMs to get ready for a technological shift which is already happening. Local models are at level with previous generation frontier models.
Qwen4.8 27B model + Hermes or OpenClaw allows to build performing and correct code, reason with it , plan.. it is going to be slow, it will start looping because of smaller context, you are going to need capable hardware, but this is just part of the process.
I am really excited to the idea that in 5/10 years time we are all going to have a personal local AI!
commit ae5c73d
I think it is already possible or it should be possible, to route an agent to either a cloud model or local model based on the effort/complexity of the task.
But I also hope that soon running local models will get a bit more affordable, right now is quite expensive.