
TC, whose current project materials use the Thinker name, evaluates compact LoRA reasoning adapters submitted by miners against a fixed base model.
Miners publish LoRA adapters for a fixed base model, and validators score them on math, long-context QA, and multiple-choice reasoning tasks before setting rewards from the resulting scores.
The project documentation describes compact adaptation of a shared frozen base model. The subnet remains registered, but its official dashboard currently shows no miners, no validators and no current-round participation, so it is not an actively operating competition today.
Bittensor miners and model researchers building compact reasoning adapters for a shared base model.
The competition evaluates small adapter artifacts against a fixed base across several reasoning tasks rather than rewarding operators for hosting a standalone inference endpoint.