You don't have to pay for model access to write code or build your own tool; thirty-one providers offer a permanently free tier. But "free" means three different things here, and the most-wanted use case (Claude Code) actually costs money. Below: the underlying pattern, five providers worth starting with, and where each one will trip you up.
All thirty-one providers, with their URLs and current limits, live in the repo below, updated daily. If the list is all you came for, go ahead; you don't need to read the rest.
What follows is the part the list doesn't tell you: which one to actually start with, where "free" stops being free, and which URL will send you into a wall on your first try.
Most of these providers speak the same format as OpenAI. In practice: whatever tool you use (Cursor, Codex, Aider, your own script), you're always asked the same three things: URL, key, model name. Once that clicks, all thirty-one providers become the same door with a different address.
Three steps
Pick a provider
New to this? Groq. No credit card, fast, and a comfortable limit.
Get the key
Most only want an email. Takes under a minute.
Paste it into your tool
URL + key + model name. Whatever the tool, it's these three.
In Python that's three lines. The code below connects to Groq; change the base_url line and the same code works with any provider.
from openai import OpenAI
client = OpenAI(
base_url="https://api.groq.com/openai/v1",
api_key="GROQ_API_KEY",
)
response = client.chat.completions.create(
model="llama-3.3-70b-versatile",
messages=[{"role": "user", "content": "Merhaba!"}],
)
print(response.choices[0].message.content)Three different things, and mixing them up leads to disappointment: some are genuinely free forever (Groq, Gemini, Cerebras), some want identity verification (NVIDIA wants a phone number), and some keep the free tier narrow and gate real usage behind payment (OpenRouter).
The real limit isn't model quality, it's requests per minute. Plenty for chatting; but write a script that calls the model in a loop and you'll hit the wall within the first minute.
What it asks, what it gives
| Provider | What it asks | Free limit |
|---|---|---|
| Groq | Email only | 30 req/min |
| Google Gemini | Email only | 15 req/min · 1,500/day |
| Cerebras | Email only | 5 req/min |
| NVIDIA NIM | Phone verification | Up to 40 req/min |
| OpenRouter | Registration | Narrow free tier |
Don't trust the list blindly
The source list is generated automatically and refreshed daily, and that's the good part. The bad part: a few rows are broken. xAI is listed twice, and Cline's URL field is empty. Model names also change often. URLs are stable, model lists aren't; always take the model name from the provider's own dashboard.
Noticeably faster responses than the rest, and no credit card, just an email. 30 requests a minute, plenty for chat and normal development. Make your first attempt here; once the pattern clicks, the others are just a different URL.
https://api.groq.com/openai/v1Take the model name from the console; Groq's free model list changes from time to time, don't rely on a fixed name.
Gives you a 1M-token context window on the free tier, the most generous option here for feeding it long documents or a codebase. No credit card. The 1,500-requests-a-day cap is invisible for most personal use.
https://generativelanguage.googleapis.com/v1beta/openai/Careful: the source list ends this URL at /v1beta; that's Gemini's own format and it won't work with the OpenAI SDK. Google's own docs say to append /openai/. The URL above is the correct one.
The provider with the most models here, and it goes beyond text: image, audio and video models too. If you want to try many different models with one key, this is the place.
https://integrate.api.nvidia.com/v1No credit card, but it does want phone verification; if you came for "no card" and don't want to hand over your number, this is where you stop. It has the most models among the card-free options; the price is your number.
Runs on its own hardware, so token generation is very fast. No credit card. You feel the difference on anything that needs an instant answer.
https://api.cerebras.ai/v1The cap is tight: 5 requests a minute. Fine for chatting, but a script making back-to-back calls will jam immediately. Speed yes, volume no.
One key gets you models from dozens of providers; you call the free ones by appending :free to the model name. Saves you from signing up with each provider separately.
https://openrouter.ai/api/v1The most-asked use case costs money here: running Claude Code against Anthropic models through OpenRouter requires a one-time $10 top-up. The free tier exists, but the Claude Code path isn't free; filing it under "free" is misleading.
Installing these tools on your own is one thing; actually building with them is another. The community has people using these daily and people building systems from scratch.
If you want a system that actually runs in your business, let's talk for 10 minutes; I'll look at what you're trying to build and tell you which path fits. Free, and not a sales pitch.
Book a 10-minute callIf you'd rather learn this alongside people doing the same work instead of on your own, the community is always open:
Join the communityThis is an affiliate link.