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

TL;DR

This study reveals that lower listed API prices for reasoning language models often do not correspond to lower actual inference costs, due to variability in token consumption, highlighting the need for cost-aware model selection.

Contribution

First systematic analysis showing pricing reversal phenomenon in reasoning models, emphasizing the unreliability of listed prices as cost indicators and the importance of cost-aware strategies.

Findings

21.8% of model pairs show cost reversal despite price differences

Removing thinking token costs reduces reversals by 70%

Per-query token consumption varies up to 9.7x, indicating irreducible noise

Abstract

Developers and consumers increasingly choose reasoning language models (RLMs) based on their listed API prices. However, how accurately do these prices reflect actual inference costs? We conduct the first systematic study of this question, evaluating 8 frontier RLMs across 9 diverse tasks covering competition math, science QA, code generation, and multi-domain reasoning. We uncover the pricing reversal phenomenon: in 21.8% of model-pair comparisons, the model with a lower listed price actually incurs a higher total cost, with reversal magnitude reaching up to 28x. For example, Gemini 3 Flash's listed price is 78% cheaper than GPT-5.2's, yet its actual cost across all tasks is 22% higher. We trace the root cause to vast heterogeneity in thinking token consumption: on the same query, one model may use 900% more thinking tokens than another. In fact, removing thinking token costs reduces ranking reversals by 70% and raises the rank correlation (Kendall's $\tau$ ) between price and cost rankings from 0.563 to 0.873. We further show that per-query cost prediction is fundamentally difficult: repeated runs of the same query yield thinking token variation up to 9.7x, establishing an irreducible noise floor for any predictor. Our findings demonstrate that listed API pricing is an unreliable proxy for actual cost, calling for cost-aware model selection and transparent per-request cost monitoring.