The Price Reversal Phenomenon: When Cheaper Reasoning Models End Up Costing More

A systematic study of 8 frontier reasoning language models reveals that cheaper API pricing often leads to higher actual costs due to variable 'thinking token' consumption. The research found that in 21.8% of model comparisons, the cheaper-listed model actually costs more to operate, with cost differences reaching up to 28x.

• →Listed API prices are unreliable indicators of actual inference costs for reasoning language models.
• →21.8% of model-pair comparisons showed pricing reversals where cheaper models cost more in practice.
• →Thinking token consumption varies dramatically between models, with some using 900% more tokens than others on identical queries.
• →Cost prediction is inherently difficult due to up to 9.7x variation in thinking token usage across repeated runs.
• →Removing thinking token costs from calculations reduces ranking reversals by 70% and improves price-cost correlation significantly.