A model, in Gonka's vocabulary, is an AI model that has been registered on chain through governance — not just deployed by a participant on its own. Registration matters economically as well as operationally: each model carries its own units-of-compute-per-token figure, set at registration, which is the per-model constant used to price every inference request served by that model. A heavier model burns more units per token, so serving it earns a participant proportionally more per token processed.

As of early July 2026, the on-chain registry lists three models: MiniMax M2.7, Kimi K2.6, and GLM 5.2. Participants are assigned to serve specific models at each epoch boundary, alongside the rest of that boundary's settlement work (weight updates, validator rotation, reward payout, vesting).

GNKScan's models page tracks what is currently registered and being served, and the inference page shows which models each active participant is serving. For how a model's pricing constant feeds into the fee formula, see Gonka Tokenomics; for the governance and registration angle, see the Gonka FAQ.