Free LLM API: Which Providers Are Still Free in 2026 (Limits, Keys and Catches)
Table of contents:
Quick answer: Yes, a free LLM API still exists in 2026. As of November 16, 2026, Google Gemini, Groq, OpenRouter, Mistral, Cloudflare Workers AI, NVIDIA, Cohere and Z.ai offer a standing free tier with rate limits. Cerebras, Anthropic and Alibaba Cloud give credits that run out instead. GitHub Models, a popular free option, shut down on July 30, 2026.
Best for most people: Google Gemini's free tier for capable models, Groq for speed, OpenRouter for many free models behind one key.
No signup at all: run an open-source model locally with Ollama or LM Studio.
The catches: low daily limits, models that change without notice, and some free tiers may use your prompts for training.
Not for production: free tiers are built for testing. Plan a paid path before you launch.
Looking for a free LLM API to build or test with?
There are still good options, but the list has changed a lot this year.
Some popular free tiers have closed, others have quietly become one-time credits, and many "free LLM API" lists online haven't caught up.
This guide covers every free LLM API worth using right now, checked against each provider's own documentation: what's free, the limits, the data catch, and how to get a key.
It also shows how to call several free providers from one code path, and what to do when free limits run out.
For that last step, ScriptRun's OpenAI compatible API gives you one key for OpenAI, Claude and Gemini models.
What counts as a free LLM API
"Free" means three different things, and mixing them up is how projects break a month after launch:
-
A standing LLM API free tier: free, rate-limited access that doesn't expire. This is what most people mean by a free LLM API.
-
Free LLM API credits: a one-time balance, often requiring a card, that runs out or expires.
-
Free open-source models you run yourself: no API fees at all, but you supply the hardware.
Free LLM APIs compared
Here are the free LLM APIs available as of November 16, 2026, grouped by type.
| Provider | Type | What's free | Card needed? | Free-tier data used for training? |
|---|---|---|---|---|
| Google Gemini API | Standing free tier | Flash and Flash-Lite models, such as Gemini 3.8 Flash; per-model limits shown in Google AI Studio | No billing needed | Yes (Google: "content used to improve our products") |
| Groq | Standing free tier | Open-weight models with per-model limits per organization | No (needed only to upgrade) | No |
| OpenRouter | Standing free tier | Models ending in "": 20 requests/min, 50/day, or 1,000/day after buying $10 of credits | No | Depends on the serving provider |
| Mistral | Standing free tier | "Free mode," the default for new accounts; lowest limits | No | Yes, unless you opt out |
| Cloudflare Workers AI | Standing free tier | 10,000 Neurons per day across models | Not stated | No |
| NVIDIA API Catalog | Standing free tier | Trial access to hosted models; no numeric limits published | Not stated | Not stated |
| Cohere | Standing free tier | Trial key: 1,000 API calls per month; not for production | Not stated | Not stated (opt-out exists) |
| Z.ai (GLM) | Standing free tier | Three Flash models priced free | Not stated | No |
| Cerebras | Credits | $5 one-time credit, expires 30 days after activation | Yes | No |
| Anthropic (Claude) | Credits | Small starter credits; amount not published | Not stated | No |
| Alibaba Cloud (Qwen) | Credits | 1 million tokens per model, valid 90 days (Singapore region) | Not stated | No |
| GitHub Models | Retired | Shut down on July 30, 2026 | — | — |
| OpenAI, DeepSeek, xAI, Hugging Face (free accounts) | No free tier | Pay-as-you-go only | — | — |
"Not stated" means the provider's pages don't say. It doesn't mean "no."
Free tiers change often, so check the linked source before you rely on a number.
Free LLM API providers with a standing free tier
#1: Google Gemini API
Google's Gemini free tier is the most capable free LLM API for most people.
Free input and output cover the Flash and Flash-Lite models, including Gemini 3.8 Flash.
Google says the free tier is "for developers and small projects getting started with the Gemini API."
Your exact per-model limits appear in Google AI Studio, not in the public docs.
The catch: Google's pricing page marks free-tier content as "used to improve our products," while paid-tier content isn't. Source: Gemini API pricing.
#2: Groq
Groq offers a free plan on its very fast inference hardware, with open-weight models such as gpt-oss-120b and Qwen.
Limits apply per organization and per model; Groq's docs show sample limits of around 30 requests per minute and 1,000 per day on popular models, and your exact limits are on the Limits page in the Groq console.
Source: Groq rate limits.
#3: OpenRouter
OpenRouter gives you dozens of models behind one key, and the ones whose IDs end in "
" cost nothing per token.Every account gets 20 requests per minute on free models, with a daily cap of 50 requests, rising to 1,000 a day once you've bought at least $10 of credits.
The free model list changes often, and whether your prompts may be used for training depends on which provider serves the model.
Source: OpenRouter limits.
#4: Mistral
New Mistral accounts start in "Free mode," with API access and no credit card. It has the lowest limits, shown only in your admin panel.
By default, Mistral may use free-mode inputs and outputs to train its models; you can switch this off in the Admin panel's privacy settings.
#5: Cloudflare Workers AI
Cloudflare gives 10,000 "Neurons" per day for free, a unit that covers different models at different rates.
It's a good fit if your app already runs on Cloudflare.
#6: NVIDIA API catalog
The NVIDIA free LLM API on build.nvidia.com offers trial access to many hosted models through NVIDIA NIM.
NVIDIA doesn't publish numeric limits, and trial use may be limited by instances or duration.
#7: Cohere and Z.ai
Cohere's trial key allows 1,000 API calls a month and isn't for production.
Z.ai lists three GLM Flash models as free.
Free LLM API credits that run out
These providers give you free LLM API credits rather than a free tier.
They're useful for a short evaluation, not for a side project you want to keep running.
-
Cerebras: no longer has a free tier. You get a one-time $5 credit after adding a valid payment method, and it expires 30 days after activation. Many lists still describe Cerebras as "free, 1 million tokens a day," which is out of date. Source: Cerebras pricing.
-
Anthropic (Claude): new accounts may get small starter credits. There's no free Claude API tier.
-
Alibaba Cloud Model Studio (Qwen): 1 million tokens per model, valid for 90 days, in the Singapore region.
Providers that no longer offer a free LLM API
GitHub Models is gone. It was one of the most popular free options, but GitHub retired it on July 30, 2026.
The playground, model catalog, inference API and bring-your-own-key endpoints are no longer available to anyone.
If your code still calls it, it will fail.
Source: GitHub changelog.
The OpenAI API, DeepSeek API and xAI's Grok API have no free tier for text generation, and Hugging Face Inference Providers have no free allowance for free accounts.
Free LLM API alternative: open-source models you run yourself
Asking whether there are any open source LLM API keys available?
You don't need one.
Open-source models such as Llama, Qwen, Mistral and gpt-oss are free to download, and local servers give them an API on your own machine:
-
Ollama: one command to download and serve a free LLM model, with an OpenAI compatible API at http://localhost
/v1/. -
LM Studio: a desktop app with a local server at http://localhost
/v1. -
vLLM: a high-throughput server for your own GPUs, at http://localhost
/v1 by default.
There are no rate limits and your data stays on your machine.
The trade-off is hardware: larger models need a strong GPU, and small models on a laptop are slower and less capable than hosted ones.
Which is the best free AI LLM for your project
There's no single best free AI LLM; it depends on what you're building.
Whether you searched for a free API LLM, free LLMs with API access or a free AI models API, these picks cover most cases:
-
Prototyping a chatbot or app: Google Gemini's free tier, for the most capable free models.
-
Low-latency features: Groq, for very fast open-weight models.
-
Comparing many models: OpenRouter's "
" models, behind one key. -
Private or offline work: a local open-source model with Ollama.
How to get a free LLM API key
Getting a free LLM API key takes a few minutes with most providers:
-
Pick a provider from the table above. Gemini, Groq and OpenRouter need no card.
-
Sign up with an email, Google or GitHub account on the provider's console.
-
Create an API key in the keys or API section, and copy it right away. Most consoles show it only once.
-
Store it in an environment variable, such as GROQ_API_KEY, never in your code.
-
Check your limits on the provider's limits page before you build on it.
A warning about "free API keys for AI models" shared online: keys posted on GitHub, forums or Discord belong to someone else.
Using them breaks provider terms, can be revoked at any time, and may expose your prompts.
Get your own free key instead, and keep your own LLM API keys private.
Use several free LLM APIs from one code path
Most free LLM API providers, including Gemini, Groq and OpenRouter, offer an OpenAI compatible API.
So one client can call all of them, and you can fall back to the next provider when one hits its limit. This makes a free LLM API for testing much more reliable:
import os
from openai import OpenAI
# Free-tier providers, in order of preference
PROVIDERS = [
{
"base_url": "https://generativelanguage.googleapis.com/v1beta/openai/",
"key": "GEMINI_API_KEY",
"model": "gemini-2.5-flash"
},
{
"base_url": "https://api.groq.com/openai/v1",
"key": "GROQ_API_KEY",
"model": "openai/gpt-oss-120b"
},
{
"base_url": "https://openrouter.ai/api/v1",
"key": "OPENROUTER_API_KEY",
"model": "openrouter/free"
},
]
def ask(prompt):
last_error = None
for p in PROVIDERS:
if not os.environ.get(p["key"]):
continue # Skip providers without an API key
client = OpenAI(
base_url=p["base_url"],
api_key=os.environ[p["key"]],
max_retries=0
)
try:
r = client.chat.completions.create(
model=p["model"],
messages=[{"role": "user", "content": prompt}]
)
return r.choices[0].message.content
except Exception as err:
# Try the next provider if this one fails
last_error = err
raise RuntimeError(f"All configured providers failed: {last_error}")
print(ask("Explain an API rate limit in one sentence."))
This is a simple form of the routing an LLM router does for you in production.
Free LLM APIs: what's the catch
-
Your data may train models. Google's free tier and Mistral's Free mode may use your prompts unless you're on a paid plan or opt out. Don't send customer or confidential data through a free tier.
-
Limits are low and shared. Daily caps run out fast in agents or batch jobs, and limits usually apply per account or organization, not per key. You'll hit 429 errors; our guide to OpenAI API rate limits explains how to handle them.
-
Free models change without notice. Providers add and remove free models, and whole free tiers disappear, as GitHub Models and Cerebras' free tier did this year.
-
They aren't built for production. Most providers say free tiers are for testing and prototyping, with no uptime guarantees.
When a free LLM API isn't enough
Free tiers are perfect for learning and prototypes.
Once real users depend on your app, you need limits you control, models that don't disappear, and a clear data policy.
|
If cost is the main concern, our upcoming guide to the cheapest LLM APIs compares paid prices per token.
FAQs about Free LLM API
Is there a free LLM API?
Yes. As of November 2026, Google Gemini, Groq, OpenRouter, Mistral, Cloudflare Workers AI, NVIDIA, Cohere and Z.ai all offer a free LLM API tier with rate limits. You can also run open-source models locally for free.
Which LLM API is free to use?
For most projects, start with Google's Gemini API free tier for capable models, Groq for fast responses, or OpenRouter's ":free" models if you want many models behind one key.
Which LLM gives a free API key?
Google AI Studio (Gemini), Groq, OpenRouter, Mistral, Cohere and NVIDIA all give a free API key after sign-up. Gemini, Groq and OpenRouter don't ask for a card.
How do I get a free LLM API key?
Sign up on the provider's console, create a key in the API keys section, copy it once, and store it in an environment variable. See the step-by-step section above.
Are there any open source LLM API keys available?
You don't need a key for open-source models. Download them and run them with Ollama, LM Studio or vLLM, which give you a local API for free. Hosted open-weight models on Groq or OpenRouter need your own free key from that provider.
Which AI models are free?
Free hosted options include Gemini Flash and Flash-Lite models, open-weight models on Groq and OpenRouter (such as gpt-oss and Qwen), Mistral's models in Free mode, and GLM Flash models on Z.ai. Open-source models are free to run yourself.
Is Google's LLM free?
Google's Gemini API has a free tier on Flash and Flash-Lite models, with limits shown in Google AI Studio. Free-tier content may be used to improve Google's products; paid-tier content isn't.
What is an LLM API?
An LLM API lets your code send text to a large language model over the internet and get a response back, usually as JSON. You pay per token, or use a free tier within its limits.
Related posts
Cheapest LLM API 2026: Verified Prices per 1M Tokens
Which is the cheapest LLM API? Verified prices for OpenAI, Claude, Gemini, DeepSeek, Qwen, Mistral and Groq, plus real workload costs and hidden fees.
AI Agent Marketplace: 10 Best Places to Find, Run and Sell AI Agents in 2026
What an AI agent marketplace is, 10 top platforms compared on cost to try and who can sell, a safety checklist, and how to list your own agent in 2026.
OpenAI Compatible API: What It Is, Who Supports It and How to Switch Safely
An OpenAI compatible API guide: base URLs for 15 providers, what breaks when you switch, a copy-paste smoke test and a safe migration checklist.
OpenRouter Alternatives: 7 LLM Gateways Compared on Routing, Pricing and Lock-In
OpenRouter alternatives compared: fees, routing, BYOK and self-hosting across 7 LLM gateways, with a migration checklist and verified 2026 prices.








