Top 8 Free AI APIs in October 2026

A free AI API is an LLM endpoint you can call with an API key and pay $0 for, inside a quota the provider publishes. As of October 22, 2026, eight providers offer a free tier with no card: Google Gemini, Groq, OpenRouter, Mistral, Cloudflare Workers AI, NVIDIA’s API catalog, Cohere and SambaNova. Six more give starter credits instead.

Every limit below comes from the provider’s own docs, read on October 22, 2026. If you want open-weight models you can self-host, read our guide to free open-source LLM APIs instead; this post covers hosted provider free tiers only.

The 8 free LLM APIs at a glance

Every base URL in the table speaks the OpenAI chat completions format, so one client covers all eight.

What counts as a real free tier

A real free tier has three properties: you pay $0, you don’t need to add a payment method, and the quota renews (daily or monthly) instead of running out once. A one-time credit grant is a trial, not a free tier. Credit routes get their own section below.

1. Google Gemini API

The Gemini API has the widest free model range here. The Gemini pricing page lists “Free of charge” free-tier input and output for Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 2.5 Pro, 2.5 Flash, the Gemma 4 family and Gemini Embedding 2, among others.

What’s not free: “Gemini 4 Argon is not available on the Free tier,” and neither are the image, Veo or Lyria models. Google no longer prints per-model RPM and RPD numbers on its rate limits page; your active limits show in AI Studio, apply per project (not per key) and reset at midnight Pacific.

The catch is data use. Google’s terms say unpaid quota content can be used to improve its products and read by human reviewers, so don’t send sensitive data. Our step-by-step guide shows how to get a free Gemini API key.

2. Groq

Groq publishes the clearest free-tier numbers. Its rate limits page shows, under the Free Plan tab:

• openai/gpt-oss-120b: 30 RPM, 1K requests/day, 8K tokens/min, 200K tokens/day
• openai/gpt-oss-20b and qwen/qwen3.8-27b: the same limits
• whisper-large-v3: 20 RPM, 2K requests/day for speech-to-text

Groq’s billing FAQ asks for a payment method only “to upgrade from the Free tier to the Developer tier.” The 8K tokens-per-minute cap is the one you’ll hit first.

3. OpenRouter free models

OpenRouter routes to many providers, and any model ID ending in :free costs nothing. Its limits doc sets 20 requests per minute and 50 requests per day for accounts that have bought less than 10 credits in total. Buy at least 10 credits once and the daily cap rises to 1,000.

On October 22, 2026, the public models endpoint listed 18 :free IDs, including google/gemma-4-31b-it:free, nvidia/nemotron-3-super-120b-a12b:free and poolside/laguna-s-2.1:free. The list rotates monthly, which is why we keep a separate roundup of free AI models on OpenRouter.

4. Mistral (Free mode)

Mistral calls its free tier “Free mode.” The quickstart states: “API access is enabled by default with no credit card required. Usage and rate limits apply.”

Mistral doesn’t publish the numbers. You’ll find yours under Admin Panel, API, Limits. The docs also list a 402 Payment Required error for “No payment method on account,” so some models or volumes may need a card.

5. Cloudflare Workers AI

Cloudflare gives every account “10,000 Neurons per day at no charge,” on both the Free and Paid Workers plans, per its Workers AI pricing page. Neurons are Cloudflare’s compute unit: @cf/meta/llama-3.1-8b-instruct uses 25,608 Neurons per million input tokens and 75,147 per million output tokens, so the free allocation buys roughly 130K output tokens a day on that model.

Text generation allows 300 requests per minute, and the allocation resets at 00:00 UTC. Some newer models (several Kimi, GLM and DeepSeek builds) require a paid billing method.

6. NVIDIA API catalog

NVIDIA’s docs say: “Members of the NVIDIA Developer Program have free access to NIM API endpoints for prototyping.” Join the free program, open a model on build.nvidia.com and generate a key.

NVIDIA staff say the old credit system was removed and the request rate “will vary per model queried.” No official RPM figure is published; the “40 RPM” you’ll see quoted comes from user posts, not NVIDIA. Production use needs an NVIDIA AI Enterprise license, so treat this as a prototyping route.

7. Cohere (trial key)

Cohere creates a trial key automatically when you sign up. Its rate limits doc says trial keys “are limited to 1,000 API calls a month,” with 20 chat requests per minute per model, 2,000 embed inputs per minute and 10 rerank requests per minute.

The restriction: trial keys “are not permitted to be used for production or commercial purposes.” The embed and rerank endpoints make it a good free route for testing retrieval, and the compatibility API runs command-a-plus-05-2026 through the OpenAI SDK.

8. SambaNova Cloud

SambaNova’s rate-limit docs define the Free Tier as “applied when there is no payment method linked with your account.” Each listed model, including Meta-Llama-3.3-70B-Instruct, gpt-oss-120b, DeepSeek-V3.2 and gemma-4-31B-it, gets 20 RPM, 20 requests per day and 200,000 tokens per day.

The plans page reads differently: its Free card says “Add a payment method and purchase credits to run your first requests.” Twenty requests a day suits a smoke test, not a prototype.

Free AI API keys with starter credits

These are trials or need a card, so they miss the main list:

• Cerebras: $5 in credits that expire 30 days after they’re granted; “New accounts receive $5 in free credits after adding a verified payment method.”
• Vercel AI Gateway: $5 per month on a subset of “Free Tier” models; a valid payment method is required before the credits work.
• Fireworks AI: a $1 starter credit, at 10 RPM with no payment method on file.
• Nebius Token Factory: $1 trial credit valid for 30 days; billing setup requires a bank card.
• Anthropic: “New users receive a small amount of free credits to test the API.” No amount is published. We cover the details in how to get free Claude API access.
• Scaleway Generative APIs: the first 1,000,000 tokens are free, with a validated payment method.

Checked and not free today

Four common answers no longer hold:

1. GitHub Models: “As of July 30, 2026, GitHub Models has been fully retired.”
2. Hugging Face Inference Providers: free users get no monthly credits; PRO users get $2 a month.
3. Together AI: requires a $15 minimum credit purchase.
4. xAI and OpenAI: both run on prepaid credits. OpenAI’s “Free” usage tier is a spending limit, not a credit grant.

Call any free LLM API with one client

Since all eight speak the OpenAI format, swap two values and keep your code:

import os
from openai import OpenAI

PROVIDERS = {
    "groq": ("https://api.groq.com/openai/v1", "GROQ_API_KEY", "openai/gpt-oss-120b"),
    "gemini": ("https://generativelanguage.googleapis.com/v1beta/openai/", "GEMINI_API_KEY", "gemini-3.8-flash"),
    "openrouter": ("https://openrouter.ai/api/v1", "OPENROUTER_API_KEY", "google/gemma-4-31b-it:free"),
}

base_url, key_env, model = PROVIDERS[os.environ.get("PROVIDER", "groq")]
client = OpenAI(base_url=base_url, api_key=os.environ[key_env])

reply = client.chat.completions.create(
    model=model,
    messages=[{"role": "user", "content": "Reply with the word ok."}],
)
print(reply.choices[0].message.content)

Wrap the call in a fallback: on a 429, move to the next provider in the dict. Our guide to handling “rate limit exceeded” errors covers backoff and retry headers.

Let your coding agent test them with Apidog CLI

Free tiers drift: a model leaves OpenRouter’s free list or a daily cap drops. Rather than retest by hand, let your coding agent keep a test case per provider and re-run it through Apidog CLI. Set it up once:

npm install -g apidog-cli@latest
apidog login --with-token <YOUR_TOKEN>
npx -y skills add https://apidog.com   # pick your agent when prompted

Then give your agent (Claude Code, Codex, Cursor or OpenCode) one prompt:

Use Apidog CLI on project <id>. Create one environment each for Groq, Gemini
and OpenRouter with the base URLs from this article and the key in
{{API_KEY}}. Add POST /chat/completions with a one-line prompt and model set
to {{MODEL}}. Create a test case that asserts status 200 and a non-empty
choices[0].message.content, run it in each environment, and report which
provider failed and the status code it returned.

The commands it runs, checked against --help on Apidog CLI 2.2.7:

apidog environment create "Groq free" --project <id> --base-url https://api.groq.com/openai/v1
apidog cli-schema get endpoint-create
apidog endpoint create --project <id> --file chat-completions.json
apidog cli-schema get test-case-create
apidog cli-schema validate test-case-create --file case.json
apidog test-case create --project <id> --file case.json
apidog run --test-case <caseId> --project <id> -e <envId> -r cli,json

A 429 or 403 in the report flags a changed free tier before your users notice. Put the same apidog run ... -r junit on a daily schedule in CI, as shown in our Apidog CLI CI/CD guide, or read how the agent workflow runs in Apidog CLI with Claude Code. Apidog’s Free plan covers up to 4 users.

Which free AI API should you pick?

• Most free models and context: Gemini, if your data isn’t sensitive.
• Clearest published limits and fast inference: Groq.
• Trying many models with one key: OpenRouter :free.
• Embeddings and rerank for RAG: Cohere’s trial key.
• Running inside Cloudflare Workers: Workers AI.

For more ways to run models without paying, see how to use LLMs for free.

FAQ

Is there a completely free AI API? Yes. Google Gemini, Groq, OpenRouter’s :free models, and Mistral Free mode all let you call an LLM at $0 without adding a card, within published or dashboard-shown limits checked on October 22, 2026.

Which free LLM API has the highest limits? Groq publishes the highest fixed numbers for one model: 1,000 requests and 200K tokens per day on openai/gpt-oss-120b. OpenRouter reaches 1,000 requests a day only after a one-time 10-credit purchase.

Do free AI API keys expire? Free tier keys don’t expire, but quotas reset daily or monthly. Trial credits do expire: Cerebras credits last 30 days and Nebius credits 30 days.

Can I use a free LLM API in production? Cohere’s trial keys and NVIDIA’s catalog are explicitly not for production, and Gemini’s free tier lets Google use your content. The other caps are too low for real traffic.

Is the OpenAI API free? No. OpenAI’s docs grant no free API credits; its “Free” usage tier is a spending limit.

Is GitHub Models still a free LLM API? No. GitHub retired GitHub Models on July 30, 2026, including the playground, catalog and inference API.