i am already on api tokens for the chinese open source models and no subscriptions. these are all available in the original form open source and priced above the inference cost. i think this is the long term option.

zero degradation in speed or quality seen.