Omron

Omron is a verifiable inference network on Bittensor subnet 2, maintained by Inference Labs. It addresses the verification problem in decentralised AI: validators either trust miner outputs or rerun the inference, and rerunning is too costly and slow. Omron has miners attach a zero-knowledge proof to each prediction, compiling the model into a circuit that attests the inference really ran on the specified model with the given inputs.

Validators only check the proof, cutting verification from seconds to milliseconds, then score miners on proof size, response time and output accuracy before setting weights. In 2024 the subnet moved to a second version that let miners choose their own models, added accuracy scoring against a local reference model, and shipped Proof-of-Weight so validators prove they actually ran the scoring process instead of copying weights. Staking and restaking strategy recommendations were the early use case, with teams such as Benqi and Renzo using its proof service.