Sarpoy vs Stoke Chat (pay-per-token)
Pay an LLM operator per token consumed
Stoke Chat and similar pay-per-token chat platforms route payment from the user to the operator running the model. Both meter access per unit, so they share the pay-per-call surface — but the economic shape differs. Stoke is a marketplace: the operator delivers tokens, the user pays. Sarpoy is a metered game: a creator funds an escrowed pot, solvers pay a rising on-chain fee to probe, and one solver wins the pot. The fee flows to the pot, not to an operator.
Choose Stoke Chat (pay-per-token)
Pick Stoke when you're trying to monetise an LLM as a service — you have a model, you want to charge for tokens, and your audience is consumer-style.
Choose Sarpoy
Pick Sarpoy when the model is a puzzle gatekeeper, the prize is the creator-funded pot, and the goal is to crack the objective rather than to consume the output.
Feature-by-feature
| Feature | Sarpoy | Stoke Chat (pay-per-token) |
|---|---|---|
| Money flow | Solver → treasury PDA → winning solver | User → operator |
| Settlement | On-chain SOL, atomic with verification | Off-chain or chain-of-choice; varies |
| Incentive shape | Single-shot pot, one winner | Continuous token sales, many users |
| Prize source | Creator-funded escrow | There is no prize; just a service |
| Verification | submit_solution gates settlement | Implicit — tokens delivered = paid |
Technical notes
Stoke's mental model: monetised inference. Sarpoy's mental model: settled adversarial game.
You can stack them: a Stoke-style inference layer could host the LLM that powers a Sarpoy bot. They live at different parts of the stack.
Stoke users want long sessions and many tokens. Sarpoy solvers want few messages and a fast solve. The UI optimisations diverge accordingly.