Epoch AI: cost of a given AI performance level falling 47% per quarter, the fastest in tech history
dl_weekly · x · 2026-10-01
Epoch AI's report "The plunging price of thought" quantifies the historic collapse in AI inference costs.
Key findings
- Since 2023, the cost of achieving a given level of AI performance has fallen 47% per quarter, or about 13x per year — faster than any other transformative technology: 4x faster than DNA sequencing, 6x faster than compute, 18x faster than lithium batteries, and 54x faster than electricity (in the century up to 1973).
- Coarser evidence suggests the price of thought has been falling at least this fast since commercial LLM inference began with GPT-3's full release in November 2021.
Variation by domain
- Game-based puzzles see slower declines (39–43% per quarter); math problems fall fastest (50–52% per quarter).
SOTA effect
- Costs drop fastest right after a capability debuts as SOTA: across five benchmarks, freshly-SOTA performance falls 66% per quarter (75x/year), halving to 32% per quarter (4.7x/year) two years later.
Data and code are open-sourced on GitHub, with interactive plots available.
More from Infra
- tandem.cpp v0.3.0: 2-6x faster prefill for models too big for VRAM, Maya DeepSeek-V4.1 support — smflx · 2026-11-16
- Developers poll real costs of running coding agents on remote VMs for work — radim11 · 2026-11-16
- Datadog's new gateway returns 402 when a model's budget runs out—so agents just retry — YvesMulkers · 2026-11-16
- Max-quality local Minimax H3 video on a 4090: community-tested int8 setup — newaccount47 · 2026-11-16
- Local V2V outfit transfer on a 16GB M3 MacBook Air: 15.7 min cut to 2.4 min — might_thinker · 2026-11-16
- FlashStates hits 3.14x vLLM prefill throughput for hybrid-architecture agentic inference — 机器之心 · 2026-11-16