Skip to content
unerrBook 30 minutes

AI Price Index · Report

The unerr AI Price Index — August 2026

A monthly index of published AI model list prices, built from unerr’s own panel of 2,886 model SKUs. This edition covers the 35 months from Sep 2023 to Jul 2026.

Published
Updated
Panel
2,886 model SKUs · Sep 2023 – Jul 2026
Unit
model-months

What did we find?

Eight findings, one sentence each. Four are ours. Four are published by other researchers and are labelled with who measured them — we do not restate anyone else’s population or period as our own.

Measured by unerr · our panel

  • Across unerr’s panel of 2,886 AI model SKUs observed monthly over 35 months (Sep 2023 – Jul 2026), 0.82% of model-months carry a change in published list price.

    Our panel: 2,886 model SKUs, 35 months

  • In unerr’s panel of 2,886 AI model SKUs over 35 months (Sep 2023 – Jul 2026), a live SKU keeps its published list price in 99.2% of months.

    Our panel: 2,886 model SKUs, 35 months

  • Of the list-price changes that do occur in unerr’s panel of 2,886 AI model SKUs over 35 months (Sep 2023 – Jul 2026), 42% are increases.

    Our panel: 2,886 model SKUs, 35 months

  • Across unerr’s panel of 2,886 AI model SKUs over 35 months (Sep 2023 – Jul 2026), about 80% of apparent aggregate price decline is new models entering the panel, not existing models getting cheaper — independently replicated on an unrelated dataset by Demirer, Fradkin, Tadelis and Peng (NBER WP 34608, about 85%).

    Our panel: 2,886 model SKUs, 35 months

Measured by NBER · not our panel

  • Demirer, Fradkin, Tadelis and Peng (NBER WP 34608) measure a within-model log-price trend of about −0.3%/year, statistically indistinguishable from zero, over the panel and period they publish.

    Demirer, Fradkin, Tadelis & Peng, NBER WP 34608

  • Demirer, Fradkin, Tadelis and Peng (NBER WP 34608) attribute about 85% of the measured aggregate decline in AI model prices to composition — new models entering — rather than to existing models repricing.

    Demirer, Fradkin, Tadelis & Peng, NBER WP 34608

  • Demirer, Fradkin, Tadelis and Peng (NBER WP 34608) find that closed-source model prices change in 1.8% of provider-model-weeks, and in 0.9% of provider-model-weeks when weighted by usage.

    Demirer, Fradkin, Tadelis & Peng, NBER WP 34608

Published elsewhere · not our panel

  • Gundlach et al. (arXiv:2511.23455) put the fall in the cost of a fixed capability level at 5–10× per year — a rate per year, not a total over any window.

    Gundlach et al., arXiv:2511.23455 · robustness check: Cottier, Snodin, Owen & Adamczewski, Epoch AI, 12 Mar 2025

What does the index show?

Two series on one logarithmic axis. Neither slope is ours: the matched-SKU trend is published by NBER, and the quality-adjusted band is a published range compounded across the window. Our panel supplies the frequency counts around them — how often a published price moves at all.

AI Price Index · Sep 2023 – Jul 2026

Index · 100 = Sep 2023 · log scale

Matched SKUs — the models you already runMeasured · NBER WP 34608−0.3%/yr−0.8% over 35 months
Quality-adjusted — the models you’d have to switch toMeasured · Gundlach et al.5–10× cheaper/yr96× to 681× over 35 months
AI price index by month, Sep 2023 – Jul 2026. Index 100 = Sep 2023.
MonthMatched-SKU index (100 = Sep 2023)Quality-adjusted index, 5× per year (100 = Sep 2023)Quality-adjusted index, 10× per year (100 = Sep 2023)
Sep 2023100100100
Oct 202310087.482.5
Nov 202399.976.568.1
Dec 202399.966.956.2
Jan 202499.958.546.4
Feb 202499.951.138.3
Mar 202499.844.731.6
Apr 202499.839.126.1
May 202499.834.221.5
Jun 202499.829.917.8
Jul 202499.726.214.7
Aug 202499.722.912.1
Sep 202499.72010
Oct 202499.717.58.25
Nov 202499.715.36.81
Dec 202499.613.45.62
Jan 202599.611.74.64
Feb 202599.610.23.83
Mar 202599.68.943.16
Apr 202599.57.822.61
May 202599.56.842.15
Jun 202599.55.981.78
Jul 202599.55.231.47
Aug 202599.44.571.21
Sep 202599.441
Oct 202599.43.50.825
Nov 202599.43.060.681
Dec 202599.32.670.562
Jan 202699.32.340.464
Feb 202699.32.050.383
Mar 202699.31.790.316
Apr 202699.21.560.261
May 202699.21.370.215
Jun 202699.21.20.178
Jul 202699.21.050.147

The models you already run barely reprice. What falls is the price of models you’d have to switch to.

Repricing is rare — a dated jump, not a trend: 0.82% of model-months carry a list-price change, and 42% of the changes that do occur are increases.

Matched SKUs — measured. Within-model price trend ≈ −0.3%/year, statistically indistinguishable from zero (Demirer, Fradkin, Tadelis & Peng, NBER WP 34608). Our panel of 2,886 model SKUs over 35 months shows the same rigidity from the other side: a live SKU keeps its price in 99.2% of months.

Quality-adjusted — measured, published, not ours. The cost of a fixed capability level falls 5–10× per year (Gundlach et al., arXiv:2511.23455); the band is that published range compounded across the window. A second estimate puts the median near 50× per year (Cottier, Snodin, Owen & Adamczewski, Epoch AI, 12 Mar 2025) — we chart the conservative one. The vertical axis is logarithmic: the two series span three orders of magnitude.

How is the index built?

unerr’s panel holds 2,886 model SKUs, observed monthly for 35 months, Sep 2023 – Jul 2026. The unit of observation is the model-month. The panel measures how often a published list price moves; it does not estimate a price trend.

What is a model-month?

One SKU observed in one calendar month is one model-month: one SKU, one month, one published list price. A model-month carries a change when that SKU’s published list price differs from the month before. Only live months count — a SKU contributes model-months for the months it is actually offered.

So 0.82% and 99.2% are shares of model-months, not shares of SKUs and not shares of price changes. 42% is a share of the changes themselves.

Why is our unit not the same as NBER’s?

We count model-months. Demirer, Fradkin, Tadelis and Peng (NBER WP 34608) count provider-model-weeks. Different panels, different periods, different denominators: 0.82% of model-months and 1.8% of provider-model-weeks are not the same statistic, and neither converts into the other. Quote each with its own unit.

Our panel: 2,886 model SKUs, 35 months · Demirer, Fradkin, Tadelis & Peng, NBER WP 34608

What does the index not measure?

Published list prices only. Negotiated discounts, committed-use rates, and what any company actually paid are not in the panel and cannot be inferred from it. The index measures posted prices, not spend.

How should this report be cited?

unerr ( 2026 ). The unerr AI Price Index — August 2026 . Published 4 August 2026 . https://unerr.dev /ai-price-index

Findings on this page may be quoted with attribution to unerr (CC BY 4.0). We do not publish the underlying panel file, so this report ships no data download.

What changed in this revision?

Nothing yet — this is the first publication. Revisions are quarterly. A revision changes numbers: if no number changed, the date does not move.

  1. First publication. Panel: 2,886 model SKUs, 35 months (Sep 2023 – Jul 2026). Four findings measured by unerr; four carried from the named published sources.

The number this feeds

The index says how often a published price moves. It does not say what your FY27 line should be — that takes your own consumption, your own contracts, and two weeks.

30 minutes. No deck, no obligation.