Line
The Price of Reaching the Same Score
Cheapest AI cost per task to reach a fixed benchmark score, indexed to the first month = 100
Plays the chart through its history. Pauses are where the data set a record or moved most.
Palanor Foresight™ comes with a free Palanor account. Create one · Continuation is open to everyone.
A level is the share of the way from guessing to a perfect score. Each line is the lowest cost per task of any model released to date that reaches that level, from Epoch AI’s own benchmark runs at list prices. Source: Epoch AI, “The plunging price of thought” (CC BY 4.0).
What it costs to get the same benchmark score from AI, month by month. Every line starts at 100 in the first month all four were measured; a line that falls to 10 means the same score now costs a tenth as much. Each drop is a newly released model that reached the score more cheaply. Costs count the tokens a model actually spends, so reasoning models are not flattered by a low price per token.
What this chart reads
- Cheapest AI Cost to Score 75% on GPQA Diamond (Epoch AI)as “GPQA Diamond, 75% level”latest 2026-10-20
- Cheapest AI Cost to Score 75% on OTIS Mock AIME (Epoch AI)as “AIME-style math, 75% level”latest 2026-10-20
- Cheapest AI Cost to Score 25% on FrontierMath Tiers 1-3 (Epoch AI)as “FrontierMath, 25% level”latest 2026-10-20
- Cheapest AI Cost to Score 25% on Chess Puzzles (Epoch AI)as “Chess puzzles, 25% level”latest 2026-10-20
Sources
Cheapest AI Cost to Score 75% on GPQA Diamond (Epoch AI). Source: Epoch AI, "The plunging price of thought" (Luke Emberson and David Roodman, 2026), epoch.ai/publications/the-plunging-price-of-thought. Data: github.com/droodman/inference-cost (data/cost_truncated_curves.csv and Epoch's model registry), linked from the report as its data. Licence: "Epoch AI’s work is free to use, distribute, and reproduce provided the source and authors are credited under the Creative Commons Attribution license." Palanor computes the daily cost frontier with Epoch's own method, which reproduces the report's Figure 2. Refreshed weekly by Palanor's epoch-sync job.
Cheapest AI Cost to Score 75% on OTIS Mock AIME (Epoch AI). Source: Epoch AI, "The plunging price of thought" (Luke Emberson and David Roodman, 2026), epoch.ai/publications/the-plunging-price-of-thought. Data: github.com/droodman/inference-cost (data/cost_truncated_curves.csv and Epoch's model registry), linked from the report as its data. Licence: "Epoch AI’s work is free to use, distribute, and reproduce provided the source and authors are credited under the Creative Commons Attribution license." Palanor computes the daily cost frontier with Epoch's own method, which reproduces the report's Figure 2. Refreshed weekly by Palanor's epoch-sync job.
Cheapest AI Cost to Score 25% on FrontierMath Tiers 1-3 (Epoch AI). Source: Epoch AI, "The plunging price of thought" (Luke Emberson and David Roodman, 2026), epoch.ai/publications/the-plunging-price-of-thought. Data: github.com/droodman/inference-cost (data/cost_truncated_curves.csv and Epoch's model registry), linked from the report as its data. Licence: "Epoch AI’s work is free to use, distribute, and reproduce provided the source and authors are credited under the Creative Commons Attribution license." Palanor computes the daily cost frontier with Epoch's own method, which reproduces the report's Figure 2. Refreshed weekly by Palanor's epoch-sync job.
Cheapest AI Cost to Score 25% on Chess Puzzles (Epoch AI). Source: Epoch AI, "The plunging price of thought" (Luke Emberson and David Roodman, 2026), epoch.ai/publications/the-plunging-price-of-thought. Data: github.com/droodman/inference-cost (data/cost_truncated_curves.csv and Epoch's model registry), linked from the report as its data. Licence: "Epoch AI’s work is free to use, distribute, and reproduce provided the source and authors are credited under the Creative Commons Attribution license." Palanor computes the daily cost frontier with Epoch's own method, which reproduces the report's Figure 2. Refreshed weekly by Palanor's epoch-sync job.
Charts are drawn from Palanor's signal library and recomputed from source data on every view. Transformations (changes, indexing, year-on-year rates) are Palanor's.