How to size Provisioned Throughput capacity for MiniMax M3
Source
x.com
Author
MiniMax (official)
Date
Context
MiniMax's July 17 guide explains why request count alone cannot size reserved M3 capacity on Together AI. Fresh input, cached input and generated output consume capacity at different rates. Its Provisioned Throughput guide therefore sizes a workload by separating those traffic types, converting each into capacity units and adding them.
The underlying commitment is to reserved processing capacity over time. Together's documentation explains that output-heavy traffic consumes units faster, cached reads consume them more slowly, and unused capacity is not refunded. The economics consequently depend on both traffic mix and how consistently the reservation is used.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
