Inference prices, and what mining could take off them
Leading open-weight models: what standard API providers charge for input and output tokens, and what those prices could become if the provider mined Pearl on the same GPUs and passed the proceeds on, at five levels of overhead. Live: the PRL price, the network's hashrate and the day's issuance are today's.
What mining does to the price
How the discount moves with the PRL price
Market prices, and the prices mining could fund: today’s PRL price
$ per million tokens. Market is the median across standard API providers, input and output each on its own. The estimates apply the discount mining could fund to both, at each overhead; red is higher than market, where mining costs the provider more than it earns.
The same at half the PRL price
Everything as above, with mining paid at half today’s PRL price. The market prices do not move; only what mining can fund does.
The same at twice the PRL price
Everything as above, with mining paid at twice today’s PRL price.
Per GPU-hour, where the discount comes from
What one GPU earns serving each model, what the miner earns on it, and what is left to pass on after the throughput the miner costs and the electricity it adds.
The models and the hardware
What each model needs to serve it, and where the throughput figure comes from.
Sources
Where to next
Related pages that answer the questions this one raises.