What a credit costs.

The median bundled rate is about 1.2¢ per credit, but that figure travels nowhere: the middle half of plans price a credit anywhere from 0.4¢ to 9.5¢, and the extremes run from $0.00016 to $25 (excluding Bolt, the rate outlier). A credit is not a unit. What is comparable is how a product prices against itself — and there the patterns bite.

Topping up costs more than the bundle

Of the 84 plans where both the bundled rate and the top-up price are public, 49 charge more per credit to top up than the plan’s own bundle — a median premium of +18%. Run low before your reset and the marginal credit is routinely dearer than the one you already paid for. Premiums reach +500% on the cheapest tiers, where the gap between the advertised bundle and the real cost of running over is widest. Only 24 of 84 plans make the top-up cheaper than the bundle; the other 11 price it identically — so even at best, topping up never beats the plan you already bought.

Top up, or upgrade? A rule of thumb

Every credit-metered buyer hits it: you’re running low — buy a top-up, or jump a tier? Top-ups are for spikes; the bundle is for sustained use. The decision comes down to a few numbers — the $-distance to the next tier, the break-even (the extra credits per month at which upgrading and topping up cost the same — below it top up, above it upgrade), and how that tier prices its credits. The last column is the tell: what a credit bought via top-up costs versus the same credit bought by upgrading. Above 1×, the upgrade is the cheaper way to buy credits — so top up only for the occasional spike, never as a habit. Below 1×, the next tier is mostly selling you seats or features (not credits), so a top-up is the cheaper patch and the upgrade only earns its price on the extras.

monday and Midjourney sit highest — their next tier is a great credit deal, so routinely topping up is the expensive choice. Where it’s below 1× (Figma, Gamma), you’re paying the upgrade mostly for seats, not credits — there, a top-up is the rational patch.

The value of a credit floats with the model

In 38 of the 75 products, one credit buys a different amount of work depending on which model it runs. A fixed dollar peg means nothing if a “premium model” quietly costs 5× the credits of a standard one. In most (25) you choose the model, so the burn is yours to control; where the product auto-routes, the credits-per-action is the vendor’s to set — and to change. That’s the structural root of credit burn-rate repricing (see What’s changing).

Where the published ladder ends

31 of the 75 products stop publishing prices above some tier and switch to “contact us.” The median public-pricing ceiling — the highest number you can read before the page goes dark — is $120/mo. The wall comes earliest for incumbents (8 of 11) and support/CX (5 of 7); coding tools publish the whole ladder (2 of 11 go dark).

What it costs in actual dollars

Every figure above is a ratio or a percent. In plain dollars, the price of entry varies as much as the credit rate: the cheapest published plan in the panel is a few dollars a month; the dearest public tier runs into the thousands. The range per category is the floor-to-ceiling of what it takes to play.