---
title: "Free APIs for Developers and Agents: What the Free Tiers Actually Give You"
url: https://apyhub.com/blog/free-apis-for-developers-and-agents-what-the-free-tiers-actually-give-you
author: ApyHub
published: 2026-08-24T15:37:54.801587Z
---

# Free APIs for Developers and Agents: What the Free Tiers Actually Give You

## Introduction

A free API is one you can call without paying, but that covers at least four different arrangements: genuinely free public endpoints, free tiers that need a card on file, trials that expire, and freemium plans that bill you the moment you cross the line. The differences only matter once you have built something on one.

This is not another list of free APIs. It is an audit of what the word actually buys you on the platforms developers reach for first, what happens when you hit the ceiling, and a second question that did not exist two years ago: what "free" means when the caller is an agent rather than a person.

## What Changed

Two things.

Free tiers used to be evaluation budgets. You made a few calls, confirmed the API worked, and upgraded. Now teams assemble products from ten or twenty services at once, and the free tier is doing real work in prototypes, side projects, and internal tools that quietly reach production.

And the caller changed. An agent composing a workflow does not know in advance which APIs it needs. It discovers them at runtime. That makes discovery itself a cost, and it is a cost nobody was measuring when free tiers were designed for humans clicking around a marketplace.

## What Makes a Free API Tier Actually Useful

Four things worth checking before you build on one.

1. **Whether a card is required.** This is the sharpest dividing line and the least advertised. A free tier that needs a card is a paid tier with a discount at the start.
2. **What the quota is in units you can convert.** "500 requests" and "1,000 credits" are not comparable until you know what one call costs. A quota you cannot translate into your own workload is not a quota, it is a number.
3. **What happens at the ceiling.** Some platforms stop serving. Some charge overage. The second one is the surprise, and it is why the card question matters.
4. **Whether it covers the whole catalog or one API.** Per-API free tiers do not pool. Ten free tiers across ten vendors means ten quotas, ten keys, and ten renewal dates.

## The Platforms

### 1. RapidAPI

The largest marketplace by catalog size, with tens of thousands of listed APIs across effectively every category.

RapidAPI is also the clearest illustration of why "free API" is an ambiguous phrase, because their own documentation splits listings into four types. **Free APIs** need no card and no subscription, capped at 1,000 requests per hour and 500,000 per month. **Pay Per Use** APIs have no subscription fee but do require a card. **Freemium** APIs have a free tier and still require a card, which RapidAPI explains is so they can charge overages if you exceed the free quota. **Paid** APIs need both a subscription and a card.

Their documentation walks through the consequence plainly: a freemium plan with 500 monthly requests and a one cent overage fee, used 600 times, produces a one dollar charge. Small, and entirely the point. The free plan billed you.

**Where it shines**

* By far the widest selection, including long-tail APIs nothing else lists
* The genuinely free category is generous at 500,000 monthly requests
* One account and one key across every API you subscribe to

**Where it lags**

* Four different meanings of "free" on one platform, distinguishable only by reading each listing
* Freemium free tiers require a card and can bill you
* Quotas are per API and do not pool, so ten APIs means ten separate limits
* Catalog size comes at the cost of curation, and quality varies sharply between listings

### 2. APILayer

A curated marketplace of production-focused APIs across currency, geolocation, email validation, aviation, and NLP.

APILayer's free plans are lifetime rather than trial, and their FAQ confirms no card is needed to start one. The complication is that there is no single free tier to evaluate, because pricing sits on each product page. The spread is wide: the User Agent API's free plan allows 10,000 monthly requests, while the NLP API's free plan allows 100. Same platform, same word, a hundredfold difference.

**Where it shines**

* Free for lifetime, no card, on most products
* Manually curated, so the average listing quality is high
* One account and one key across the portfolio

**Where it lags**

* No unified pricing page, so evaluating cost means visiting each product
* Free quotas range from useful to nominal with no obvious logic
* Per-API subscriptions that do not pool, so a multi-API product means several plans
* Overages on paid plans are billed after the period ends rather than blocked

### 3. Eden AI

An aggregation layer over AI models and providers, covering OCR, speech, translation, sentiment, image generation, and more.

Eden AI runs a freemium model with a free tier that does not require a card. Their own published comparison places them alongside Google AI Studio, Groq, Cerebras, and Mistral as providers offering genuine cardless free access. What they do not publish clearly is a single headline free quota, so the practical advice is to check current limits on their pricing page before depending on it.

**Where it shines**

* Free tier with no card, which is not the norm in AI APIs
* One integration reaches many underlying model providers
* Side-by-side comparison of providers on latency and cost is a real differentiator

**Where it lags**

* The free allowance is not stated as a single clear figure
* Scope is AI only, so it does not cover conversion, validation, or general utility work
* Abstracting over providers hides some provider-specific controls
* Routing through a middle layer adds latency

### 4. Public API Lists

The GitHub-hosted directories of free public APIs, which rank at or near the top for most "free API" searches.

These are not platforms. They are curated link lists pointing at endpoints run by whoever runs them, and that is both the appeal and the problem.

**Where it shines**

* Genuinely free, frequently no authentication at all
* Excellent for learning, tutorials, and throwaway prototypes
* No account, no signup, no commercial relationship

**Where it lags**

* No SLA, no support, no uptime commitment, and nobody to email
* Links rot, and an endpoint can vanish without notice
* No compliance posture, which rules them out of anything touching user data
* Quality and maintenance vary enormously between entries

### 5. ApyHub

A curated catalog of 450+ services and 1,500+ endpoints, growing continuously, covering file conversion, validation, extraction, AI, geolocation, SEO, and more, under one subscription.

The [free Starter plan](https://apyhub.com/pricing) requires no card and gives you 1,000 atoms a month at 5 requests per second, with one API key. Atoms are the pricing unit: a measure of the work a call performs rather than a count of calls made, in the same way tokens measure work for a language model. Most catalog endpoints cost 1 atom, so 1,000 atoms is roughly 1,000 calls. Heavier endpoints cost more. [SERP Rank Checker](https://apyhub.com/apyhub/service/serp-rankings-for-keyword) is 250 atoms, which makes it four calls a month on the free plan. There is a fuller explanation of the unit below, since it is the thing that makes the rest of the model work.

The structural difference is that the quota is not per API. Those 1,000 atoms are spendable across the entire catalog, so testing eight different services during a prototype draws from one pool rather than eight separate free tiers.

**Where it shines**

* No card, and the free tier reaches every service in the catalog rather than one
* Atoms pool across APIs, so headroom is not stranded per vendor
* One key, one bill, one operational model regardless of how many services you call
* Every service carries machine-readable certification covering GDPR, SOC 2, and ISO 27001 alignment
* Every endpoint is MCP-ready by default

**Where it lags**

* 1,000 atoms is modest, and on expensive endpoints it is a handful of calls rather than a working allowance
* One API key and one seat on the free plan, so team evaluation needs a paid tier
* The catalog is curated rather than exhaustive, so genuinely long-tail APIs are more likely to exist on a larger marketplace
* Paid entry starts at $16 per month, which is real money against platforms with a permanently free path

## Side by Side

|                              | Free quota                            | Card required | At the ceiling       | Quota scope   |
| ---------------------------- | ------------------------------------- | ------------- | -------------------- | ------------- |
| RapidAPI (free listings)     | 1,000/hr, 500K/mo                     | No            | Blocked              | Per API       |
| RapidAPI (freemium listings) | Set per API                           | Yes           | Overage billed       | Per API       |
| APILayer                     | 100 to 10,000/mo depending on product | No            | Upgrade prompted     | Per API       |
| Eden AI                      | Not published as one figure           | No            | Upgrade prompted     | Per account   |
| Public API lists             | Whatever the host allows              | No            | Undefined            | Per endpoint  |
| ApyHub                       | 1,000 atoms/mo                        | No            | Blocked until top-up | Whole catalog |

Two columns do most of the work here. The card column tells you whether "free" can become a bill. The scope column tells you whether the quota survives contact with a real project that uses more than one API.

There is a third thing the table cannot show, because four of the five platforms measure in the same unit and that unit is the problem.

## The Unit Problem: Why Request Counts Don't Reflect Work

Every free tier above except one is denominated in requests. 500 requests, 10,000 requests, 500,000 requests.

A request is not a unit of work. It is a unit of counting.

Sending a two hundred byte currency lookup and sending a twelve megabyte scanned document for OCR both cost you one request. One of those took a database read. The other took compute, storage, and an AI model. Charging the same for both is not pricing, it is accounting convenience, and it produces two predictable outcomes: cheap calls subsidise expensive ones, and vendors compensate by splitting heavy capabilities into separate products with separate plans.

The AI industry already solved this. Nobody bills LLM usage per request, because a three word prompt and a fifty page document are obviously not the same transaction. They bill per token, in and out, so the bill tracks the work. That shift happened fast and nobody argues about it now.

API pricing largely never made that jump.

**What atoms are.** Atoms are ApyHub's unit of work, playing the same role tokens play for models. Every endpoint in the catalog is priced in atoms according to what a call actually does. Most utility endpoints cost 1 atom: [Validate IBAN](https://apyhub.com/apyhub/service/validate-iban), [Convert Word to PDF](https://apyhub.com/apyhub/service/convert-word-to-pdf), [Extract Text from PDF](https://apyhub.com/apyhub/service/extract-text-from-pdf). Heavier ones cost more, because they are heavier. [SERP Rank Checker](https://apyhub.com/apyhub/service/serp-rankings-for-keyword) is 250 atoms, and it costs 250 because a live search query is not a string operation.

**Where it gets more precise.** For some APIs the work is not knowable until the request is processed, so a fixed price would be wrong in one direction or the other. Providers can price those dynamically, returning the actual cost of the individual request. A Word to PDF conversion can be priced at a base plus a rate per page, so a ten page document costs 70 atoms and a hundred page document costs 250. A text to speech endpoint can price by model, so the standard voice costs 50 and the premium voice costs 100. An AI endpoint can combine model cost with input and output processing, exactly as a model provider would.

The part worth noticing is what happens when a request turns out to be cheaper than expected. The gateway charges a base amount up front, compares it against the real cost once the provider responds, and **refunds the difference** if the actual work came in lower. A call priced at 50 that only did 30 atoms of work is billed at 30, not 50. Request counting has no mechanism for that, because it has no concept of a request being smaller than another one.

**Why this matters for a free tier specifically.** A 1,000 atom free allowance sounds smaller than 500,000 free requests, and on a marketplace where every listing has its own separate quota, that comparison is not meaningful anyway. What atoms give you is a number you can convert. You can look at an endpoint, see what it costs, and know how far your allowance goes before you write anything. On a request-counted platform you know how many calls you get and nothing about whether the calls you need are the cheap kind or the kind the vendor put behind a separate plan.

The unit is also what makes cross-catalog pooling possible at all. Headroom can only be spendable across a currency conversion, an OCR job, and an AI inference call if those three things are measured in something comparable. Request counts cannot pool honestly, because pooling them would mean pretending they represent the same thing.

## What "Free" Means for an Agent

This is the part the older comparisons do not cover, and it is a different question with a different answer.

When an agent uses APIs, discovery has a cost that has nothing to do with the vendor's price list. Kaiwalya Koparkar at Gravitee [measured this recently](https://www.linkedin.com/pulse/your-ai-agent-loads-200-tools-use-12-youre-paying-all-koparkar-yzzwf): connecting an agent to GitHub's MCP server hands it 116 tool definitions, which serialize to roughly 34,940 tokens sitting in context before the agent does anything useful. A code review agent needs about seven of them. In a realistic multi-server setup spanning five services, he counts over 200 tool definitions against roughly 12 actually needed.

That overhead is charged on every turn. Prompt caching discounts the re-read but does not eliminate it, and his measured results after curating the tool surface were a 67% drop in per-request cost and a 64% drop in token load.

So for an agent, a free tier that hands over a whole catalog of tool definitions is not free at all. It is a token bill paid to a model provider instead of an API vendor.

[ApyHub's MCP server](https://apyhub.com/mcp) is built around this. Rather than loading every endpoint as a tool definition, the agent connects once and searches the catalog at runtime, retrieving only the handful of definitions it needs for the task in front of it. Searching costs nothing. Atoms are only spent on calls actually made.

There is a second benefit worth naming, because it is the opposite of the obvious fix. Trimming a tool list per request breaks prompt caching, since the cached prefix has to stay byte-identical to be reused, and a list that changes per request shatters one warm prefix into several cold ones. A single stable search tool does not have that problem. It is the same bytes every turn.

[**Explore the catalog →**](https://apyhub.com/catalog)

## When Free Is Genuinely Enough

Free tiers get a bad name from people who expected production capacity. Used for what they are, they are fine.

Free is enough for learning an API's shape, for a hackathon, for an internal tool with a handful of users, for the prototype stage of anything, and for confirming that a service actually returns what its documentation claims before you commit to it.

Free is not enough the moment real users depend on it, the moment you need an SLA, the moment the data involved has a compliance posture attached, or the moment failure has a cost. At that point the question stops being what a plan costs and becomes what happens when it breaks.

## Conclusion

There is no single best free API platform, and the answer depends on what you are building.

For breadth and long-tail coverage, RapidAPI is unmatched, provided you read each listing carefully enough to know which of the four "free" arrangements you are signing up for. For a small number of specific, well-defined utilities, APILayer's lifetime free plans are hard to argue with. For AI capability specifically, Eden AI's cardless free tier is genuinely unusual in that category. For learning and throwaway work, public API lists are exactly right and nothing else is cheaper.

For a project pulling on several different capabilities at once, or one where an agent is doing the calling, the thing to look at is not the size of the free tier but whether it pools and whether discovery is free. Those are the two properties that decide whether a free plan survives your second week.

[**Try ApyHub free →**](https://apyhub.com/)

## FAQ

**What is a free API?** A free API is one you can call without paying, but the term covers several arrangements: public endpoints with no authentication, free tiers requiring an account, freemium plans that need a card on file, and trials that expire. Checking which one you have matters before you build on it.

**Which free APIs do not require a credit card?** APILayer's free plans, Eden AI's free tier, ApyHub's Starter plan, and RapidAPI's listings marked specifically as free all work without a card. RapidAPI's freemium listings do require one, which their documentation explains is so overages can be charged.

**What is the best free API for testing?** For testing an integration end to end, a free tier that covers many services beats a large quota on one. ApyHub's free plan spends 1,000 monthly atoms across the whole catalog, so you can test several different capabilities from a single pool and key.

**Are there free public APIs with no API key?** Yes. The GitHub-hosted public API directories list many endpoints that need no authentication at all. They are good for learning and prototypes, and unsuitable for anything with an SLA, support, or compliance requirement, since there is no commercial relationship behind them.

**How do I get free API keys?** On most platforms you create an account and a key is issued immediately. ApyHub, APILayer, and Eden AI issue keys without a card. Check whether the key covers the whole catalog or a single API, since that determines how many keys you end up managing.

**What are the best free AI APIs?** Eden AI offers a cardless free tier across many model providers, and ApyHub includes AI endpoints for OCR, sentiment, summarisation, translation, and chat completion within the same free allowance as the rest of its catalog. Both approaches avoid managing separate keys per model vendor.

**What are atoms and how do they compare to request counts?** Atoms are ApyHub's unit of work, playing the role tokens play for language models. A call is priced by what it actually does rather than counted as one request, so a lightweight validation costs 1 atom while a live search query costs 250. Request-based pricing charges the same for a 200 byte lookup and a 12MB OCR job.

**Can API pricing change per request?** On ApyHub, yes. Providers can return the real cost of an individual request, so a Word to PDF conversion can be priced per page and a text to speech endpoint can be priced by model. If the actual work comes in cheaper than the amount charged up front, the difference is refunded.

**What happens when a free API tier runs out?** It depends on the platform, and this is the question worth asking first. Some block further calls until the next period, some prompt an upgrade, and some bill an overage per request. The third is why a card requirement on a free tier is a meaningful signal.

**Do AI agents use more of a free tier than a developer does?** Not necessarily in API calls, but they carry a cost developers do not: tool definitions loaded into context. Measurements of a single MCP server have found over 34,000 tokens of definitions before an agent does any work. Catalogs an agent can search at runtime avoid that overhead, and searching consumes no quota.

**Is a free tier enough to run something in production?** Rarely. Free tiers work for learning, prototyping, internal tools, and evaluation. Once real users depend on a service, the relevant questions become uptime, support, and compliance posture, none of which a free plan typically covers.

***

## About ApyHub

[ApyHub](https://apyhub.com/) is a curated API catalog and trusted operational layer for developers, teams, and AI agents. One subscription covers the whole catalog, billed in atoms, with headroom pooled across every API rather than locked to individual services.

Every service carries machine-readable certification covering data handling, retention, and standards alignment including GDPR, SOC 2, and ISO 27001, so compliance can be evaluated before integration rather than discovered after it. Every endpoint is MCP-ready by default, and the MCP server lets agents search the catalog at runtime rather than loading every tool definition up front.

ApyHub is headquartered in Amsterdam, with offices in the Netherlands, Greece, and India. The catalog holds 450+ services and 1,500+ endpoints, with new APIs and providers onboarded continuously. The free tier requires no credit card.

Publishing an API? [Become a provider →](https://apyhub.com/become-a-provider)