[Step 9] Earning rewards

In the Consumption Phase, model hosts—whether using their own servers or Spectral's hosting service—are responsible for processing inference requests from Spectral system users.

Inference Requests and Rewards

Upon receiving an inference request, a model host has a Service Level Agreement (SLA) timeframe of 600 seconds to provide a response. Successfully responding within this timeframe earns the host reward points. These points are used to calculate the host's share of the revenue generated from user fees.

Fee Structure

Users are charged a fee for each inference request. The distribution of this fee is as follows:

  • 85% to model hosts who respond within the SLA.

  • 5% as a creator fee.

  • 5% as an admin fee.

  • 5% as a Spectral platform fee.

The majority of the fee goes directly to the modelers, with the most active modelers earning the most fees during each epoch period. Modelers who respond to little or no inferences in a given epoch, may be removed from the list of eligible participants.

Epochs and Earnings Withdrawal

Reward points are accumulated over epochs, which are defined as 7-day periods. Epochs work as follows:

  1. New modelers are added to the list of Top 10 modelers every Monday 11:59 pm GMT

  2. Any new modeler has to wait until Wednesday 11:59 pm GMT of the following week to be able to use their model for providing inferences to consumers

  3. This mechanism allows for modelers to be eligible to provide inferences for at least 1 epoch in length, even if a more performant modeler has been discovered.

  4. Now, once a modeler has deployed their model, their activity from Thursday 12:00 am GMT to Wednesday 11:59 pm GMT, every week, will be eligible for claiming rewards

  5. Rewards from Thursday of the previous week to Wednesday of the current week (as described above) will be available for claiming on the Thursday of the current week. At the end of each Thursday, Modelers may withdraw their earnings, which are proportional to the number of reward points they've accumulated during that epoch. This proportion is calculated based on their share of the total fees generated by all successful inference responses within the epoch.

Last updated