The station is where a job gets created: pick a hardware tier, choose how hard the result is checked, see every part of the price, and escrow it on chain. If the job fails, the requester is made whole out of the provider's stake โ not out of somebody else's escrow.
The tier is the hardware class you need. Every tier minimum is read live from the staking contract, so this page and the staking page can never disagree.
Read straight from the two contracts. No wallet needed.
The scheduler may only pick a provider whose node is registered, in good standing, and staked at or above the job's tier minimum. The ladder is read from the staking contract at assignment time โ it cannot be argued with.
Timeout 2%, offline 5%, bad output 30%, proven fraud 100% of that provider's stake. Half of whatever is slashed goes to the requester whose job failed, 30% to the treasury, 20% is burned.
Staking OMC is a security deposit, not a yield. There is no advertised rate here and there will not be one: what a node earns depends on work it actually delivers.
The same five tiers priced on the staking page, read live from the contract.
| Tier | Hardware | Minimum stake |
|---|
Counters for the requester side, live.
The same eight steps the whitepaper describes from the requester's side, with the part each contract plays marked.
Container image, resources, region, ceiling price, verification policy and protection level.
The ceiling is locked in OMCComputeMarket. It can never be charged more than this.
Nodes that miss the VRAM, driver, region, TEE, stake or reputation bar are dropped.
Survivors are ranked on TFLOPS, VRAM, bandwidth, latency and reputation, weighted per job class.
A provider is drawn from the top-k and recorded on chain โ the eligibility check runs in that transaction.
The node pulls the image, gets the keys after assignment, runs the job and returns an attestation.
Spot-check, redundancy or TEE attestation decides whether the output is accepted.
Pass: the provider is paid, the fee splits, reputation updates. Fail: bond slashed, requester refunded.
GPU-hours ร reference rate ร verification multiplier, plus a 3% protocol fee, minus 10% of that fee when you pay in OMC. You set the ceiling; the contract refuses to settle above it.
Spot-check adds the audit rate on top. Full redundancy costs 2โ3ร because the work is done 2โ3 times. TEE carries a premium and needs TEE-capable nodes to be online.
OMC targets below the median of the centralized GPU marketplaces. Hyperscalers ran $6.88โ$12.29 per H100 SXM GPU-hour, marketplaces $2.49โ$6.16, other decentralized networks $1.25โ$3.00. The measured OMC benchmark is published after mainnet.