Open weight = verifiable inference. Audit the model, fork it, run it at home. Same logic as Bitcoin: don't trust, verify. Qwen 3.8 at 2.4T open is more interesting than any closed 10T you can't inspect.