The refill became the product

For most of the AI software on the AI Credit Index, the refill used to be the part of the bill you could not see. In June, 39% of the plan tiers on the common panel that sold extra credits did not publish what those credits cost. The price was in the app, after you had run out, when you were least likely to shop around. By October that share had fallen to 31%, and ten products had a pack price that the Index now classes as public where June's capture did not. This essay is about what happened when they did.

The short version: once a refill has a published price, it is a product, and vendors price it like one. The premium over the plan rate rose on the common panel from a median of 6% to 15%, and across the whole Edition 2 panel it sits at about 18%. The refill is no longer the awkward thing you buy when you miscounted. It is where the growth in the bill lives.

Who published, and at what#

The ten newly public prices, from the archived pages:

Pika printed a pack ladder, 60 credits per dollar on packs from $25 to $500 and 66 on the $1,000 pack, where June offered “Purchase more” with no rate. Against plan credits at 1.11 cents, the pack credit at 1.67 cents is a 50% premium. Descript published 350 credits for $35, 1,000 for $80 and 4,000 for $200. Apollo put self-serve top-up tiers on its pricing page as it unified its credit types. Atlassian Rovo named $0.01 per credit for extra usage, with default-on billing at that rate announced for December. Abacus ChatLLM stated “10K credits cost $10”. HeyGen named $0.05 per extra credit for Business users in its help centre, a rate that was in-app only in June. Instantly documented a recurring add-on of 50,000 credits for $425 a month. Zendesk’s per-resolution rates, $1.50 committed and $2.00 pay-as-you-go, appear in the pricing page’s own data. Canva priced its AI Pass add-on, US$100 per person per month, in its help docs, though the one-time top-ups it added for Business users stay in-app. The tenth, Replit, is a classification change rather than a new rate: its Starter tier now points at the help page that already priced Core and Pro packs in June.

Two more worth naming, though both were already classed as public in June: Udio’s pricing page lists “100 credits - $3.00” and “1000 credits - $25.00”, and Recraft’s credits doc sells packs at a cent a credit, 200 for $2 up to 1,600 for $16.

Against that, three products went the other way. Figma withdrew its pack tiers and per-credit rate from public pages (“at the pay-as-you-go rate in place at the time”). Ideogram de-published its pack sizes along with its free allowance. Cursor removed its dollar allowances altogether. Disclosure improved on net, not uniformly.

The premium is the price#

Here is why the published rate matters more than the published fact.

Across the 84 Edition 2 tiers where both the plan rate and a public pack rate exist, the median refill costs about 18% more per credit than the plan. But the median hides a shape. At one end, 24 of those tiers sell refills cheaper than the plan: ClickUp’s AI add-ons bundle credits at 0.66 to 0.93 cents and sell packs at a tenth of a cent; Gamma’s Plus plan bundles at 1.2 cents and refills at 0.4; HeyGen’s Business refill is half the plan rate; Kaiber’s packs are about a third cheaper than its plans. For those vendors the plan is the expensive way to buy credits and the pack is the discount, which usually means the plan is really selling something else, a seat or a feature set, with credits attached.

At the other end the refill is priced as a penalty. Udio’s packs cost 2.5 cents a credit against plan rates of 0.42 to 0.50 cents, a premium of 400% to 500%. Activepieces sells extra credits at 0.7 cents against a 0.2-cent plan rate. Gumloop’s refill is 170% over. PixVerse’s packs are double its Pro rate and two and a half times its Premium rate. Kling’s Ultra plan bundles at 0.62 cents and refills at 1.52. Midjourney’s extra GPU hours cost twice its bundled hours. Pika’s 50% on its smaller packs (36% on the $1,000 pack, which is the figure the Index records) sits in the middle of this group, and eesel’s $0.80 extra credit is 34% over its $299-for-500 plan.

Those two shapes are two different businesses. A below-plan refill says: we want you consuming, buy more, the plan was the entry fee. An above-plan refill says: the allowance is the price point, and everything above it is where we earn. Neither is wrong. But a buyer who sees “top-ups available” on both pricing pages and reads them as the same offer is misreading the one that matters.

What a published pack does to the bill#

The Pika essay worked this through for one product, so one paragraph here. When Pika published its ladder, it also changed what a video costs in credits, so a Starter subscriber who makes 35 cheapest-model clips a month went from paying exactly the $10 plan in June to paying $15.01 in September: the plan plus three 100-credit blocks priced at the pack rate, $1.67 each. The sticker was unchanged. All of the increase arrived through the refill, at the published premium.

That is the general mechanism. The allowance sets where the refill starts; the burn rate sets how fast you get there; the pack price sets what each step beyond costs. A vendor that publishes the pack price has told you the third number. Many buyers then negotiate the first, the plan price, which is the one that applies once. The pack rate applies to every credit of growth.

Why vendors are publishing now#

Three reasons, from the pattern across the ten.

The first is that the refill has become a large enough share of revenue to be worth designing. In the calculator run in the Pika essay, a user at 60 clips a month paid more in packs than in plan. Once that is true for a meaningful slice of customers, an in-app price that nobody can plan for is a revenue line with no demand curve attached. Publishing it is how you give it one.

The second is that competitors are publishing. Udio, Recraft, Pika, Descript and Apollo now all publish pack prices, Pika with the cleanest ladder. A public pack rate is a comparable number, and once one vendor in a category offers it, in-app-only pricing starts to read as evasive. Credits are how AI hides its prices described the hiding; this quarter is the first in which the Index can see competitive pressure on it.

The third is the one buyers should notice. A published price is also a published premium, and a vendor that is confident in its premium publishes it. Udio did not hide a 500% premium; it printed it. The vendors still keeping the price in the app are, on this evidence, less sure of theirs.

What to do with this#

Ask for the pack rate before the plan discount. If the vendor has published it, you have the number. If not, the Index records that tier as hidden and the hub’s transparency findings show how common that still is: a quarter of all tiers that sell a top-up. Hidden is a negotiating position, not a fact of life.

Read the premium against your growth, not your current usage. A 20% premium on a plan you never exceed costs nothing. The same premium on a team doubling its usage is the whole increase in next year’s bill.

Model the refill as a second price. The AI Pricing Calculator loads any Index product’s plan and pack rates together, so the bill at a given usage is a single run rather than a guess. The How to buy AI without overpaying checklist asks for the pack price in writing. As of Edition 2, for about three-quarters of tiers you can simply read it.

Every pack price above is on the product’s page in the Index with the archived capture it came from; the premium figures and their denominators are on the Edition 2 page.