For an AI application, the best pricing unit is usually the highest-value unit you can measure, attribute to your product, and explain to customers. Charge for tokens when customers are buying model access itself; use credits tied to recognizable work when your app completes variable tasks; and consider outcome fees when results are observable and attributable. Many products will need a deliberate mix rather than a single meter.
Choose the unit that matches what the customer buys
Start with the customer’s mental model, not your infrastructure bill: what unit of value does the buyer already understand, and can you measure it consistently? Buyers who ask about tokens may be seeking comparability or control, but that does not mean tokens are the clearest unit for every application.
As an Amazon Associate I earn from qualifying purchases.
Erten and Wang’s framework is to price at the highest layer of value that can be reliably measured, attributed, and defended. In practice, that means distinguishing raw model access from application work and from business outcomes.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
How the pricing models compare
| Model | Best fit | Customer legibility and budget control | Main pricing risk |
|---|---|---|---|
| Tokens | Customers buying model inference or access directly | Familiar to technical buyers comparing model usage; spend can be difficult to forecast when workload details vary | Customers may have to estimate context length, retrieval, retries, reasoning time, and output volume; volatile usage can make budgets and ROI comparisons harder |
| Credits tied to work | Applications completing useful but variable tasks | Clearer when credits correspond to recognizable tasks and relative effort; a shared pool can cover multiple workflows | A credit system becomes opaque if it is merely token consumption translated into another unit, or if effort bands are hard to understand |
| Outcome fees | Work with observable, attributable business results | Directly connects charges to results customers value | Attribution and result definitions must be clear enough to support stable pricing |
| Hybrid | Products with distinct access, variable-work, or unusually costly components | Can align each meter to a different layer of value | Multiple meters add explanation and billing complexity if their purposes are not explicit |
The comparison is qualitative: Erten and Wang do not publish a numerical ranking of these models. The right choice depends on the product’s workload, customer expectations, and ability to observe cost and value.
#1 Best Overall
When token pricing makes sense
Tokens are a defensible meter when the customer is buying access to inference itself and can use token counts to compare or control that access. They are less intuitive when a customer buys an application that completes work. In that setting, the customer may need to forecast not just output tokens but also context length, retrieval volume, retries, and reasoning time—implementation details that do not necessarily explain what the completed work was worth.
Token pass-through can still belong in a hybrid plan for unusually expensive or volatile model usage. If you use it, identify the specific usage it covers and explain why it is metered separately rather than making every application task appear to be priced in infrastructure units.
Make credits represent recognizable work
Credits can suit the middle layer of the AI stack: the product does useful work, but the effort and delivery cost vary from task to task. A sound credit system abstracts the underlying infrastructure while giving customers a useful sense of relative effort. For example, a small bug fix might consume fewer credits than a multi-file feature; summarizing one contract clause might cost less than reviewing a full agreement; enriching one record might use fewer credits than a multi-step account research workflow.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Credits work best when customers can understand the work they buy, the relative effort bands, and what a shared pool can cover. They do not become customer-friendly simply by replacing a token label with a credit label. If the mapping is a hidden translation of token use, buyers may lose rather than gain transparency.
Use outcome pricing only when attribution holds up
When a product can observe a meaningful result and credibly attribute it to its work, an outcome can be a more direct pricing unit than either tokens or task credits. Examples include a resolved support conversation, a qualified lead, a booked meeting, a processed claim, or dollars recovered.
Before adopting an outcome fee, define the outcome and how it is measured. If the result depends substantially on other teams, systems, or external events, attribution may be too ambiguous to support a predictable charge. In that case, a work-based credit or a hybrid meter may be easier to explain and defend.
Keep customer transparency separate from the cost meter
The customer-facing explanation does not have to expose every unit that drives your delivery cost. Show what work was completed and where capacity went; do not assume that the billing meter must match the underlying cost meter. A task history, effort band, or usage breakdown can help customers understand consumption even if the bill is based on credits or outcomes.
This distinction matters because AI delivery costs can include inference, retrieval, search, tool calls, data, retries, and failed runs. A pricing model should account for those costs without making the customer decode every internal component. Separating value from delivery cost also gives the business a clearer way to manage gross-margin exposure, though it does not by itself guarantee improved margins.
Build a hybrid without multiplying confusion
A hybrid can use seats for access, credits for variable work, token pass-through for unusually costly or volatile usage, and outcome fees where attribution is clean. Each component should correspond to a distinct layer rather than charging several meters for the same work without explanation.
Rank #4
- Seats: access to the product or workspace.
- Credits: variable application work, with recognizable effort bands.
- Token pass-through: a clearly identified category of expensive or volatile model use.
- Outcome fees: a result that can be observed and attributed reliably.
As one reported example, Andreessen Horowitz says Clay’s 2026 pricing memo separates Data Credits for third-party data from Actions for orchestration work, uses fixed pricing for predictable-cost models, and passes through actual token cost without markup for more volatile, expensive reasoning models. That description is reported second-hand in the article; the memo was not independently checked.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the reported buyer survey does—and does not—show
Erten and Wang report that 27 of 50 technical AI buyers preferred credits tied to recognizable work, while 14 preferred tokens. The article does not provide the survey’s sampling method, geography, question wording, or enough methodology to judge representativeness. Treat those counts as the authors’ reported survey, not as a universal estimate of buyer preferences.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallA practical decision sequence
- Name the purchase. Is the customer buying raw inference, completed application work, or a business result?
- Test measurement and attribution. Confirm that the proposed unit can be counted consistently and, for outcomes, tied credibly to your product.
- Check customer comprehension. Ask whether a buyer can forecast the unit and understand what it buys without estimating hidden implementation details.
- Map delivery-cost exposure. Account for inference, retrieval, tools, data, retries, and failed runs when deciding whether a fixed price can withstand workload variation.
- Choose the fewest meters that fit. Use a hybrid only when each meter serves a distinct purpose and can be explained plainly.
- Expose useful usage detail. Help customers see what work was performed and how capacity was used, regardless of the internal cost meter.
Source: Andreessen Horowitz, “You are not a model. Don’t price per token.”, by Tugce Erten and Sarah Wang, published August 27, 2026.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
