Better Insights Faster Metrics And New Developer Power Tools
Source
openrouter.ai
Author
OpenRouter
Date
Context
OpenRouter's reasoning-model performance metrics count work that occurs before the final answer. Its release note defines latency as the wait until the first reasoning tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → and includes both reasoning and answer tokens in throughput. A reasoning token is a generated text unit used during the model's intermediate reasoning. Metric changes.
These definitions change what a speed figure tells you: the latency metric marks the start of reasoning output, while throughput covers more than the final response. They help describe model activity, but should not be read as the time needed to finish an answer. Published definitions.
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.
