HomePrice per Intelligence Point and Cost per Task
How Chatty Each Model Is: Tokens Burned Running the Same Benchmark (Implied, Millions)
How Chatty Each Model Is: Tokens Burned Running the Same Benchmark (Implied, Millions)#
On the same benchmark, Sonnet 5 at max effort burns a billion tokens, 31 times the most frugal model; GPT-6 Astra is the most frugal in the $20 band.Draft, pending review
How to read this chart
Implied consumption = task cost ÷ list price, for the 17 models scoring 40 or above. Red is China, blue is the US.
Source: Published model pricing and the Artificial Analysis index; FinSight compilation · Updated 2026-09-04
↩ This chart lives on the paged3-intel-price-mon-04