GPT-5's reasoning effort setting swings token use 23x and intelligence from frontier to GPT-4.1 level. Picking the effort level matters as much as picking the model.
Key takeaways · AI-distilled
Artificial Analysis tested four GPT-5 reasoning settings in August 2025. High scored 68 on the then-current Intelligence Index, while Minimal scored 44. The endpoint name alone did not describe the tested capability.
In the August 2025 GPT-5 tests, High used about 82 million output 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 Minimal about 3.5 million. Comparing reasoning settings therefore required measuring token use as well as accuracy.
Artificial Analysis reported that GPT-5 High and Medium led its AA-LCR evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → in August 2025. AA-LCR tested reasoning over document sets totaling about 100,000 tokens, rather than context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition →-window capacity alone.
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.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
