The Price Reversal Phenomenon
Free while signed in. Answers cite the passages they came from.
The model you think is cheaper might actually cost you more. A new study systematically evaluates 8 frontier reasoning language models across 9 diverse tasks and reveals that listed API prices are misleading. In 21.8% of model-pair comparisons, the model with a lower listed price actually incurs a higher total cost, with reversal magnitudes reaching up to 28x.
Hidden thinking token costs: The root cause is vast heterogeneity in thinking token consumption. Reasoning language models generate a variable and often large number of thinking tokens that are invisible to users but billed as output tokens. On the same query, one model may use 900% more thinking tokens than another.
Concrete cost reversals: Gemini 3 Flash’s listed price is 78% cheaper than GPT-5.2’s, yet its actual cost across all tasks is 22% higher. These reversals are not edge cases but systematic patterns that affect real deployment decisions and budget planning.
High variance within single models: Even for a single model on a single query, thinking token consumption varies by up to 9.7x across repeated runs. This unpredictability makes cost forecasting nearly impossible when relying on listed per-token prices alone.
Call for transparent cost monitoring: The authors recommend that AI providers implement per-request cost breakdowns and cost estimation APIs that expose the expected thinking overhead. Without this transparency, developers are effectively making pricing decisions with incomplete information.
Get next week’s papers.
The same picks and the same summaries, in your inbox. Free, and 176 issues deep.
Subscribe on Substack