The real free tiers of Gemini, Groq, OpenRouter, Cloudflare Workers AI, Hugging Face and DeepSeek. Exact rate limits, which need a credit card, and copy-paste Python. Last verified 2026-07-08.
In 2026 the "free LLM API" world split into two camps: rate-limited free (Groq, Gemini, Cloudflare) where you pay nothing but hit a wall on volume, and credit-gated free (OpenRouter, DeepSeek, Hugging Face) where a one-time top-up unlocks far higher daily limits. For prototypes, demos, and low-volume production (a few thousand calls/day) you can run entirely free. For anything user-facing at scale, budget ~$5–20/mo.
| Provider | Free quota | Best for | Card needed? | Notes |
|---|---|---|---|---|
| Google Gemini | ~15 RPM / ~1,500 RPD on 2.5 Flash (free preview). Limits shift by model. | Long context, multimodal, quality | No | Free quota was reduced in 2026 — verify on ai.google.dev. Best free model quality. |
| Groq | ~30 RPM / ~14,400 RPD on Llama 3.3 / 4 models | Speed (lowest latency) | No | Fastest inference free. Per-model token caps. Great for chatbots. |
| OpenRouter | 20 RPM fixed; 50 req/day on :free models. $10 top-up → 1,000/day. | Model choice + failover | No (credits optional) | One API, ~30 free models, auto failover. Daily cap is the catch. |
| Cloudflare Workers AI | 10,000 Neurons/day free (reset 00:00 UTC) | In-app inference, edge | No | Runs next to your Worker — zero egress. Small models 1.5k–3k RPM. |
| Hugging Face Inference | Free rate-limited access via Inference Providers (credits vary) | Open models, experiments | No | Per-provider free quota; can be flaky under load. |
| DeepSeek | Free trial credits (promo), then paid; tight free limits | Reasoning, cheap scale | No (trial) | R1 distill is strong. Cheapest paid after trial (~$0.50/M out). |
import os, requests
# pip install google-generativeai (or raw REST below)
url = "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent"
params = {"key": os.environ["GEMINI_KEY"]}
data = {"contents":[{"parts":[{"text":"Explain free LLM APIs in 2 sentences."}]}]}
r = requests.post(url, params=params, json=data, timeout=30)
print(r.json()["candidates"][0]["content"]["parts"][0]["text"])from openai import OpenAI
client = OpenAI(base_url="https://api.groq.com/openai/v1", api_key=os.environ["GROQ_KEY"])
r = client.chat.completions.create(
model="llama-3.3-70b-versatile",
messages=[{"role":"user","content":"Hello from Groq!"}])
print(r.choices[0].message.content)from openai import OpenAI
client = OpenAI(base_url="https://openrouter.ai/api/v1", api_key=os.environ["OR_KEY"])
r = client.chat.completions.create(
model="google/gemini-2.5-flash:free",
messages=[{"role":"user","content":"Hi free tier!"}])
print(r.choices[0].message.content)// inside a Worker (wrangler)
export default {
async fetch(req, env) {
const r = await env.AI.run("@cf/meta/llama-3.3-70b-instruct-fp8-fast", {
prompt: "Summarize free LLM APIs." });
return Response.json(r);
}
}Bind an AI binding in wrangler.toml; 10k neurons/day free.
We run a $0.29 civilization — every product ships on free/serverless infra. For client chatbots we default to Groq (speed) or Gemini (quality), both no-card. For demos that live inside a Cloudflare Worker we use Workers AI so there's zero extra hosting. When a client needs scale we add a one-time OpenRouter $10 credit or move to DeepSeek paid — still a fraction of frontier-model cost.
Get our free LLM tooling updates