AI performance costs have dropped about 47% per quarter, report finds
A new Epoch AI report by Emberson and Roodman says the cost of achieving a given level of AI performance has fallen an average of some 47% each quarter over the past three years, a 13‑fold drop per year. The authors compare OpenAI’s o3 model, which cost about $0.30 per question to reach 75% on GPQA Diamond in January 2025, with GPT‑5.6 Luna, which achieved the same score for roughly $0.0004 in…
Key points
- Cost per performance level fell about 47% each quarter, a 13‑fold yearly drop, per the report.
- OpenAI o3 cost $0.30/question for 75% GPQA Diamond in Jan 2025; GPT‑5.6 Luna cost $0.0004 in mid‑2026.
- The 725‑fold cost reduction in under 18 months suggests cheaper inference for frontier models.
The report argues that cheaper inference at a fixed performance level reduces the threat from open‑source models, because frontier models become not only smarter but also dramatically cheaper to run. This rapid cost decline is presented as faster than any other transformative technology in history.
The story so far
5 episodes →- AI performance costs have dropped about 47% per quarter, report findsthis story
The Price of Intelligence Is Falling Rapidly
marginalrevolution.com · 23 September 2026
Loading the full article…
This text was published by marginalrevolution.com and written by Alex Tabarrok. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
2sources- Hacker News discussion · 75 pointsnews.ycombinator.com
- Before Trying GPT-6 Astra, Compare It with Your Own Work | 3 Tasks with Answer Keys and an Adoption Decision SheetPress · note.com ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- OpenAI hosts DevDay 2026 in San Francisco · 1 src
- Anthropic launches Claude Opus 5.5 with 40% lower costs and Fable 5.1-level performance · 15 src
- OpenAI adds GPT-6 Sol and Luna models to ChatGPT and Codex · 47 src
- Anthropic releases Claude Opus 5.5 with Fable 5.1-level performance and 40% lower cost · 78 src
- TypeSafe AI launches Jev, a decision model claimed up to 193.6x faster and 444.6x cheaper · 49 src
Comments
via GitHub Discussions