Edition 2 covers 75 products and 293 plan tiers, observed 21 Sept – 12 Oct 2026 and published 13 October 2026. The median bundled rate is about 1.2¢ per credit across 187 paid tiers (the middle half of plans sit between 0.4¢ and 9.5¢; extremes $0.00016 to $25); the median top-up premium is +18% across the 84 tiers where both prices are public; 25% of the 208 tiers that sell a top-up hide its price.
Sample definitions, the data window and every denominator are stated below. Nothing on this page is computed from observations dated after the cut-off.
Scope
Every live product of the launch panel re-read from its own pricing and help pages between 21 September and 12 October 2026, plus the products added from the October competitor screen; Relay.app, which shut down in September 2026, leaves the current view and stays in Edition 1. From this edition Bolt’s token-denominated bundled rate is excluded from the rate aggregates as an outlier (it stays on the panel and in every export row); Edition 1’s figures are unchanged.
Definitions used on this page
- Product with a published observation
- A product whose pricing page was captured to a dated, archived observation on or before 12 October 2026 and passed verification. Draft captures never enter an edition. 75 products qualify.
- Plan tier
- One published plan on a qualifying product’s latest in-window observation. 293 tiers.
- Bundled rate
- Monthly price ÷ included credits, for paid tiers that publish both. Free tiers contribute no rate. 187 tiers.
- Top-up premium
- Cheapest public top-up price per credit ÷ bundled rate − 1. Requires a public pack price on a pricing page or in help docs; in-app-only prices never enter. 84 tiers.
- Tiers that sell a top-up
- Plans marked as offering additional credits, whether or not the price is public. 208 tiers; the disclosure split is the transparency finding.
- Aggregate floor
- Any aggregate over fewer than five products is suppressed. Missing values are never imputed.
- Rate outliers
- Bolt meters individual tokens, so its bundled rate (as low as $0.0000018 per token) sits about 90 times below the next-lowest rate on the panel and nearly four orders of magnitude below the median, and it alone drove the Edition 1 whole-panel spread to 13 million to one. Excluded from rate aggregates from Edition 2 (decision 2026-10-12). The product stays on the panel for every other count and in every export row.
Key figures
Compared with Edition 1, like for like
The whole-panel figures above are not comparable with Edition 1: this edition adds 24 products and loses 1. The honest comparison is the 51 products observed in both, with every aggregate recomputed over that common panel and Bolt kept out of the rate rows on both sides (the rate-outlier rule), which is why Edition 1's column here differs from its own headline.
| Same 51 products | Edition 1 | Edition 2 |
|---|---|---|
| Median bundled rate | 1.6¢ n = 129 | 1.2¢ n = 131 −23% |
| Median top-up premium | +6% n = 38 | +15% n = 42 |
| Top-up price hidden | 39% of 137 | 31% of 144 |
Of those 51 products, 8 changed a plan's monthly price, 8 kept a price and changed what it includes (the credit burn-rate channel), and 10 published a top-up price that was hidden before. Plans are matched by their stable tier identifier, so a renamed plan still counts; a plan that appeared or disappeared does not. The dated moves are in the change log.
Essays written from this edition
Each reads the data as it stood in Edition 2; a later edition does not rewrite them.
- Pika repriced without changing a price — Pika kept every sticker price and raised every paid allowance in September 2026, yet a 720p clip costs 28% more. The arithmetic, the demand logic, the margin.
- The meter cut both ways — Runway, Luma and v0 passed falling model costs through their credit meters this quarter; Clay, Activepieces and Make made the meter less predictable.
- The refill became the product — Top-up pricing went public on ten more AI products in Edition 2, and the median premium over the plan rate rose with it. Negotiate the pack price.
- Outcome pricing arrived in support, and only in support — Fin, Zendesk, Help Scout and Gorgias price the resolution; Agentforce and eesel reached further and pulled back. Why support is where AI outcomes are billed.
- The free tier is being retired — Pika cut its free credits to zero, Ideogram stopped printing its allowance, Tasklet dropped its free plan. Free AI is turning from a product into a funnel.
- Incumbents, eighteen months in — Figma, Copilot, Notion, HubSpot, Asana and ClickUp bolted AI meters onto seat products. How each hardened, and five questions a seat buyer must ask.
- The redenomination playbook — An opinionated guide to changing an AI pricing unit without hiding it, from 75 pricing pages: what belongs on the card, in the table, in the FAQ and in help.
Post-publication corrections
None recorded.
Corrections change records inside the edition window; they are dated here and in the change log so that a figure quoted from an earlier read of this page can be reconciled.
Where to go from here
- The current view — is this edition today.
- Methodology & data — how a row is built, the source ledger, field definitions.
- All editions.