GoKawiilEpoch AI: Cost of matching AI performance fell 47% per quarter since 2023
A new Epoch AI report by Emberson and Roodman finds that the price of achieving a given level of AI performance has dropped roughly 47% per quarter over the past three years, a 13-fold annual decline. The report cites OpenAI's o3, which cost about $0.30 per question to score 75% on GPQA Diamond in January 2025, compared with a much newer model reaching similar scores for a fraction of a cent by mid-2026.
GoKawiil's interpretation of the reporting above, not reported fact.
This trend suggests that AI capability is becoming cheaper almost as fast as it is becoming smarter, which could weaken the competitive edge that open-weight or cheaper alternative models currently claim over frontier systems. If frontier labs keep cutting inference costs while improving performance, the argument that open models offer superior cost-efficiency may hold less weight over time. The pace described—faster than historical technology cost curves—implies broader economic effects as AI usage scales, though the report's authors do not specify what those downstream effects will be.
- AI performance-per-dollar has fallen about 47% per quarter over the last three years, per Epoch AI.
- An example cited shows a roughly 725-fold cost drop for similar GPQA Diamond performance between OpenAI's o3 and a later model in under 18 months.
- The trend implies frontier models are becoming both more capable and cheaper to run, potentially narrowing the cost advantage of open-weight alternatives.
Source: marginalrevolution.com — Alex Tabarrok, 2026-09-23
Published there as: “The Price of Intelligence Is Falling Rapidly”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.