Monetizing an API, often one of the final hurdles in the lengthy process of building and deploying an API, is one of those steps that feels like it should be one of the easiest. But as anyone who has gone through the process will tell you, that’s rarely the case. And recent developments in AI are making this process even trickier.

AI isn’t just changing how APIs are used — it’s prompted a wider discussion about how API providers charge their consumers. That includes a wave of new API monetization models optimized for AI agents, from token-based pricing to outcome-based models, and beyond.

As agentic API consumption increases, with agents combining calls together behind the scenes (and occasionally disregarding the API contract in the process), providers are being forced to reconsider how they position, secure, and promote their API services. Let’s take a closer look.

API Monetization Models (and How AI Is Changing Them)

Historically, when pricing an API, most developers have taken one of the following paths:

• Usage-based: Consumers are charged based on their usage volume, such as the number of calls.
• Value-based: Pricing is based on a measurable business outcome delivered by the API.
• Tiered: Different pricing tiers that bundle features or limits based on company size.
• Freemium: Basic access is free, with additional features or higher limits via upgrades.
• Subscriptions: A fixed fee, monthly or annual, grants users ongoing access to the API.

Most API pricing strategies, particularly freemium variants, rely on giving consumers a small taste (such as a 7-day trial or a limited number of calls) of the API to prove its value before they buy in.

But with agentic consumption of APIs on the rise, these may no longer be the most appropriate pricing models. An AI tool is unlikely, for example, to test out an API in the same way a human would. Higher volumes of requests (with unpredictable frequency), autonomous background consumption, and increased computational loads are all factors to consider here, too.

And for APIs that relate directly to AI, like speech-to-text APIs, text-to-speech APIs, or image generation APIs, this issue is further compounded by the fact that costs are more likely to align with criteria like tokens or GPU cycles than they are calls or requests.

So what might a world where “per call” pricing is no longer the status quo look like?

4 Newer Models of API Monetization

Quick note: It’s worth pointing out that not all of the pricing models we’ll mention here are completely brand new, or even new to the API space, but are included because we can reasonably expect to see their adoption increase as AI’s role in API consumption increases.

1. Token-Based Pricing

With tokenization, large language models (LLMs) break down blocks of text (or other types of content) into small units that they can “understand” and process. The number of tokens used in an API call that involves AI could vary considerably based on, for example, the size and complexity of the prompt provided.

Instead of trying to approximate the average number of tokens consumed during a particular API call, API providers might choose to charge based on the number of tokens used instead. This has the added benefit of encouraging the AI API’s consumers to prompt more mindfully.

ChatGPT, Arvae.ai, and Swarms all offer token-based pricing for their APIs, broken down by input versus output.

2. Hybrid Subscriptions

A hybrid model makes sense for APIs that combine standard, static methods (such as account lookups) with high-processing methods driven by AI. It can also come in handy in cases where users are unfamiliar with token-based billing or find it off-putting. In this scenario, API providers can continue to charge via their current usage-based model, with additional charges.

Beginner, intermediate, and pro packages, for example, continue as normal with the option of “AI-intensive packages” at $X per one million tokens when AI workloads spike. Where possible, API providers should include ballpark figures of how many tokens a function might consume.

Kipps.AI is a good example of a company offering a tiered pricing model based on the number of calls and agents used, with more flexible credit-based pricing for AI interactions (which includes hefty savings for using your own API keys from AI providers).

3. Agent Credits

AI agents don’t necessarily consume APIs in the same way that human developers do. They might autonomously connect multiple calls or APIs to achieve the desired response from a prompt. Forging, in other words, their own desire paths — think that well-trodden shortcut outside the paved path in your local park — through a set of APIs to assemble a suitable response.

When considering pricing that reflects these chained operations, packaging “agent credits” that can be used across different APIs and pieces of software might more accurately reflect costs than “per call” models.

AgentMark has a nice example of what this might look like in practice on their pricing page. It breaks down the process of fetching campaigns, summarizing them using GPT-4o, then generating a chart of the results, and how this would impact a monthly credit allowance.

4. Outcome-Based Pricing

When a user prompts a generative AI tool, they’re usually seeking a discrete response. Typical prompts, for example, begin with phrases like “summarize findings from…”, “create an image that…”, or “analyze this dataset to…” AI prompts that involve APIs often look like this behind the scenes, too, and it’s possible to monetize based on the end result of that combination.

A more experimental approach to API monetization involves charging for successful outcomes, rather than raw usage. It’s an interesting technique, but one that’s not without risk: it requires careful definitions of what a ‘successful outcome’ looks like, as well as in-depth calculations of how many calls, tokens, or credits it could realistically take to reach one of those outcomes.

This is something that Intercom is dabbling with right now, via their Fin AI agents. Seats are billed monthly, with an additional charge per Fin resolution. In their case, that’s defined either as a customer confirming that an AI answer resolved their issue or them not requesting more help.

The Future of Charging for APIs

Is now the time to burn your API monetization models to the ground and start all over again? In a word, no. Much has been written and podcasted on the subject of agentic API consumption, and new architectural strategies like AI gateways have emerged. But agentic consumption has not, and will not, replace human traffic overnight.

Despite everything that we’ve written above, most of the tenets and best practices associated with monetizing an API — free trials, volume discounts, rate limiting notifications, warnings when plan limits are due to be exceeded, and others — remain (and will continue to remain) the same.

Though it’s worth bearing in mind that AI agents don’t “look” at APIs the same way a human consumer might. Let’s say that an AI tool has been prompted to complete a task as cheaply as possible. If that tool happens to run across a free-to-use API with no rate limits in place, there’s a real risk that it’ll hammer the service in a way that a conscientious human consumer wouldn’t.

On the flipside, a tool that’s been prompted to prize security above everything else might turn to an API that’s much more expensive than competitors but embodies API security best practices in a way comparable services don’t. (Yes, an AI tool may become your best customer).

,Although this article focuses primarily on the impact AI might have on API pricing, the above is yet another reminder that API developers now need to consider AI and LLM tools in the same way that they do human consumers when they market, document, and secure their services.