The big picture.

A note on reading this: the panel is 51 products, observed not sampled. Everything below is association, not causation — patterns in how these products price, not claims about why.

Where credit pricing is heading

Every product here sits somewhere on one continuum: from access-priced to usage-priced — does a seat buy everything, or does usage drive the bill? The drift runs one way (the meter only ever grows, from a bolt-on toward the whole invoice), so a product’s rung is a read on its likely next move. And the shape is the finding: the panel piles up in the middle. 34 of the 51 products sit at a hard allotment — a real ceiling you hit and then pay to lift — while the access-priced left has all but emptied and the usage-priced right stays a thin frontier: only 3 have reached pure consumption. Most of the market has converged on the same answer — meter it, cap it, sell the refill — and parked there.

Rung 1 Bundled none in the panel

AI is included in the seat — no meter at all. You pay for access; usage is free at the point of use.

e.g. Granola, Wispr Flow, Obsidian — outside the index by definition (no meter to measure)

Next move Bolt on a meter once AI usage starts to bite on margins.

Rung 2 Soft meter 2 products

A meter exists but doesn’t bite — “unlimited*” with fair-use fine print, throttling rather than a cut-off.

Dream Machine (Luma), Genspark

Next move Harden the cap as compute costs climb.

Rung 3 Hard allotment 34 products

A real ceiling: run out and you’re stopped, or must buy another credit pack.

Adobe Firefly (generative credits), Apollo, Bardeen, Bolt, Canva AI (Magic Studio AI allowance), +29 more

Next move Add automatic overage so usage keeps billing past the allotment.

Rung 4 Metered overage 12 products

The meter keeps running above the allotment and auto-bills — usage becomes a line on the invoice.

Cursor, Devin, ElevenLabs, Figma AI credits, GitHub Copilot, +7 more

Next move Lean into usage: drop the seat floor and price the action directly.

Rung 5a Per-action 1 product

Usage is the price — you buy actions or credits and spend them, with little or no subscription floor.

Agentforce (Flex Credits)

Next move Tie the price to the outcome, not the action.

Rung 5b Per-outcome 2 products

You pay per result delivered — a resolved ticket, a closed task — not per action taken.

Fin, Zendesk AI agents

Next move The destination: price equals value delivered.

The archetype map

Cross transparency (the opacity composite, 0–4) with audience and four archetypes mark the corners: the honest self-serve tool that publishes its whole ladder, the hide-the-refill consumer app, the open enterprise incumbent, and the enterprise black box behind a “contact us.” But the honest read is the middle: 39 of the 51 products sit at a moderate opacity of 1–2 — mostly business and prosumer tools. The pure archetypes are real but thin; most AI pricing is an undifferentiated, moderately-opaque middle.

Who these products are built for

Tagged by go-to-market center of gravity, the panel skews to business and prosumer; pure consumer credit-metering is the minority.

Not everything meters

The meter isn’t universal — and where it’s absent is a pattern. The AI-native productivity tools we checked price flat per-seat, not by credits: Granola (AI notes, $14–$35/user, “unlimited”) and Wispr Flow (AI dictation, a free word-cap then a flat unlimited Pro) skip the meter entirely. Credit-metering clusters in generation (media, code) and agentic work — where each action carries a real, variable compute cost worth passing through — and thins out in steady, low-variance workloads, where flat per-seat is simpler to sell and to buy. The rule of thumb the market implies: if your cost per action is low and predictable, don’t meter. (Why these are excluded: methodology.)

The opacity composite

Score each product 0–4 on four opacity signals — a hidden top-up price, a “contact us” wall, model-dependent burn, and a fully unstated out-of-credits behaviour. The higher the score, the harder it is to know what you’ll actually pay. The darkest corner (3–4 of 4): Genspark, Dream Machine (Luma), Magnific AI suite.

The entry-tier penalty

The smallest plan is punished hardest when it tops up. On plans priced under $30/mo, the median top-up premium is +16%; on plans at $30+ it’s +7%. The buyer with the least room to negotiate pays the steepest marginal rate — the clearest case being a design tool whose entry tier tops up at +350%.