← The lexicon

Token throughput

n. Models new · 2026

How many tokens a system can generate per second or per dollar. Increasingly quoted as the real measure of AI infrastructure capacity and cost.

“The new chips roughly doubled token throughput per rack.”

See also

Part of the Extended list

Token throughput is on the Extended list: builder jargon, industry terms, and newer slang for when you want to go further. It entered common use around 2026.