Cacheon

Cacheon is a live, on-chain competition to build the fastest inference server for a fixed open-source model. Miners submit containerized servers and compete on speed while maintaining output correctness.

The endpoint is not a paper or a blog post. The endpoint is an inference server ranked #1 on the leaderboard, serving a top open-source model faster than vLLM on identical hardware. That is the product: the fastest way to run that model, discovered through open competition.

Product-market fit comes when the system evolves from a benchmark leader into the backend people actually choose for latency-sensitive inference. That includes agent frameworks that need fast long-context replies, Bittensor subnets that depend on reliable Qwen/GLM serving, and aggregators looking for differentiated performance. The target users are teams willing to pay for faster end-to-end responses and better GPU efficiency without building their own inference stack.