Same-performance AI costs fall ~47% per quarter, possibly faster than any tech in history
APPSO · wechat · 2026-09-24
Citing EpochAI data, APPSO analyzes a paradox: chips, power and data centers keep getting pricier, yet the cost of having AI complete equal-difficulty tasks is falling faster than perhaps any general-purpose technology ever. Across five benchmark categories over three years, cost at equal performance drops 47% per quarter (13x/year) — 4x faster than DNA sequencing, 6x than computing, 18x than lithium batteries, 54x than electricity. Example: o3 hit a GPQA Diamond level at $0.03/task-equivalent in early 2025; 18 months later GPT5.6 Luna matched it at $0.0004, a 725x drop. Costs fall fastest (66%/quarter, 75x/year) right after a capability becomes SOTA. The takeaway: frontier training still demands rising capex, but price premiums evaporate in months — models are becoming like electricity, and value shifts to workflows, data, and delivery.
Related event: Epoch AI: Cost of Equivalent AI Performance Falls 47% Per Quarter(13 posts)→
More from Infra
- Open-source ai-memory gives Claude Code, Codex, Cursor and 20+ AI coding tools a shared memory — alex_verem · 2026-10-19
- OpenRouter Data Shows US Model Token Use Flat for 8 Weeks as Chinese Providers Hit 79.2T Weekly — SumitGup · 2026-10-19
- CMU seminar digs into Floe lakehouse engine: no buffer pool, I/O faster than io_uring — sh_reya · 2026-10-19
- Data centers use just 0.85% of global primary energy, by the numbers — DarinFeinstein · 2026-10-19
- vLLM 0.32.0 adds structured-decision API reading first-token probabilities — AccBalanced · 2026-10-19
- GPU pricing is now set by cost of capital, not spot rates, argues Maya — AccBalanced · 2026-10-19