Epoch AI: cost of SOTA AI performance falls 66%/quarter, slowing to 32% after two years
FinanceYF5 · x · 2026-09-26
Epoch AI's report "The plunging price of thought" quantifies AI inference cost declines: since 2023 the cost of a given performance level has fallen 47% per quarter (13× per year) — faster than any transformative tech in history.
- 4× faster than DNA sequencing, 6× than compute, 18× than lithium batteries, 54× than electricity (century to 1973)
- By domain: game-based puzzles fall 39-43%/quarter, math problems fastest at 50-52%
- Key pattern: cost falls fastest right after a capability debuts as SOTA — 66%/quarter (75× per year) on average across five benchmarks, halving to 32%/quarter (4.7×) two years later
- The author poses a trillion-dollar question: if AI labs collectively slow model R&D, will price drops slow too
Data and code are open-sourced on GitHub.
Related event: Epoch AI: Cost of Equal-Capability AI Drops 47% Per Quarter(5 posts)→
More from Infra
- AI infra panel: 25 years of 'cheaper compute shrinks the market' predictions never came true — GregoryDiamos · 2026-09-26
- Comparing Ghent and imec silicon photonics PDs: 320 Gb/s links but 5 dB grating coupler loss — jwt0625 · 2026-09-26
- House votes 417-3 on bill making data centers pay added grid costs — VraserX · 2026-09-26
- Railway launches free no-account VMs (59 min) and OpenCode cloud agents — jasonkneen · 2026-09-26
- DistribAI v2 pools free Colab/Kaggle GPUs into ~2x RTX 5090 training power — Enderchef · 2026-09-26
- steipete pushes back on agent cost concerns: capable models now $0.1 per million tokens — steipete · 2026-09-26