Mining that does real work.
modelOS is proof-of-useful-work for the AI era. Every mined block is a verifiable model inference. Running AuxPoW, miners earn MDL and PRL simultaneously.
How it works
Proof of Useful Work
Every mined block is a verifiable AI inference — a matrix multiplication through a real language model. No wasted compute. The work that secures the network is the same work that powers AI.
AuxPoW: Two Chains, One GPU
modelOS runs AuxPoW. Mine MDL and PRL simultaneously with a single machine. One GPU. Two block rewards. Both chains secured by the same verifiable model inference.
Inference Mining
The noisy-GEMM kernel that hashes your PoW work is the same operation that runs open-weight transformer layers. Every mined block is a verifiable model inference — mathematically unified.
Miners Set Their Own Fees
inference_tx miners set the fees they charge for on-chain AI jobs. Market-driven pricing. No platform tax. The miner who wins the block fulfills the inference request and earns the fee.
Open Models, Open Source
Built on open-weight models: DeepSeek R1 from 7B to 70B. No API keys. No vendor lock-in. The protocol is fully open source — fork it, run it, improve it.
Inference Miners Are Coming
A new class of GPU operator is emerging. vLLM miners earn block rewards while idle and inference fees when jobs arrive. The first blockchain where your model earns its keep.
“The open-source Anthropic — except the miners own the network.”
No cloud. No gatekeeper. Inference is permissionless. Open-weight models, open protocol, open rewards.