Cost, Billing, and OpsOctober 12, 2026Flatkey Team

Free LLM API (2026): 10 Options Compared by Real Limits, Models, and Catches

Ten free LLM API options checked against official docs in October 2026: what is free, the exact limits, the hidden catches, and code to fall back between them.

Free LLM API (2026): 10 Options Compared by Real Limits, Models, and Catches

Yes, you can call a capable LLM API for free in October 2026: Google's Gemini API free tier, Groq's Free plan, OpenRouter's free models, Cloudflare Workers AI, and NVIDIA's hosted NIM endpoints all serve real models at zero cost, within daily caps. None of them is unlimited. The tightest numbers are public: Groq allows 1,000 requests per day on openai/gpt-oss-120b, OpenRouter caps free models at 50 requests per day (1,000 after a $10 credit purchase), and Cloudflare gives 10,000 Neurons per day. Every limit below was checked on each provider's official documentation on October 21, 2026.

This guide is for indie hackers and small teams who want to prototype without a bill, and need to know which free tier fits which job, what each one quietly costs you (data use, rate limits, model choice), and what to switch to when you outgrow it. It also flags one option most older lists still recommend: GitHub Models was fully retired on July 30, 2026.

Free LLM APIs compared (October 2026)

Provider	What is free	Published limits	Main catch	Source
Google Gemini API	Free input/output tokens on Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, 2.5 Pro/Flash/Flash-Lite	Per-model limits shown in AI Studio	Free-tier content is used to improve Google products	Gemini pricing
Groq Free plan	Hosted open models incl. `openai/gpt-oss-120b`, `qwen/qwen3.8-27b`, Whisper	30 RPM, 1K RPD, 8K TPM, 200K TPD (gpt-oss-120b)	Low tokens per minute	Groq rate limits
OpenRouter free models	25+ models with IDs ending in `:free`	20 RPM; 50 requests/day, 1,000/day after $10 in credits	Failed attempts count toward the daily cap	OpenRouter pricing
Cloudflare Workers AI	10,000 Neurons per day on Workers Free	Resets 00:00 UTC	Some frontier models require a paid plan	Workers AI pricing
NVIDIA NIM (build.nvidia.com)	"Free inference with leading models" such as Nemotron and DeepSeek V4.1 Flash	Not listed on the landing page	Meant for evaluation; limits can change	build.nvidia.com
Vercel AI Gateway	$5 of credit per month on the free tier	Per-model rate limits, lower than paid	Only free-tier-eligible models	Vercel AI Gateway pricing
Cerebras Free Trial	$5 in credits after adding a verified payment method	Credits expire after 30 days	Trial, not a permanent tier	Cerebras rate limits
Cohere trial keys	Evaluation API keys	1,000 API calls per month	Not for production use	Cohere rate limits
Ollama (local)	Free, open-source runtime on your own machine	Your hardware	You supply the GPU or CPU	Ollama
Hugging Face Inference Providers	No monthly credits for free users; PRO gets $2/month	Pay-as-you-go after credits	Not actually free for free accounts	HF pricing

All limits as published on October 21, 2026. RPM = requests per minute, RPD = requests per day, TPM = tokens per minute, TPD = tokens per day.

No longer available: GitHub Models has been fully retired since July 30, 2026; its playground, catalog, inference API, and BYOK are gone. If a tutorial tells you to use a GitHub token as a free LLM key, it is out of date.

Quick picks by job

Best overall free tier for building an app: Gemini API. It is the only option here that gives free access to a current frontier-lab model family with a 1M-token context on the Flash line.
Fastest free inference for open models: Groq Free plan, if your prompts are short (8K tokens per minute on gpt-oss-120b).
Most model variety for testing: OpenRouter free models, once you have bought $10 of credits to lift the cap to 1,000 requests per day.
Free tier inside an edge or serverless stack: Cloudflare Workers AI.
No rate limit at all: run a model locally with Ollama. You trade API limits for hardware limits.

The 10 options in detail

1. Google Gemini API free tier

Google's pricing page lists "Free of charge" input and output for Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, 2.5 Pro, 2.5 Flash, and 2.5 Flash-Lite. Gemini 4 Argon and 3.1 Pro Preview are paid only. Exact rate limits live in AI Studio, not in the docs, and Google says preview models get tighter limits (rate limits page). The catch is data: on the free tier, "Used to improve our products" is marked Yes. Keep customer data off free-tier keys. Full details in our Gemini API pricing and free tier guide.

2. Groq Free plan

Groq publishes its Free plan limits per model (rate limits). As of October 21, 2026:

Model	RPM	RPD	TPM	TPD
`openai/gpt-oss-120b`	30	1,000	8,000	200,000
`openai/gpt-oss-20b`	30	1,000	8,000	200,000
`qwen/qwen3.8-27b`	30	1,000	8,000	200,000
`whisper-large-v3`	20	2,000	audio: 7,200 s/hour, 28,800 s/day

Groq notes that cached tokens do not count toward rate limits and that limits apply per organization. The endpoint is OpenAI-compatible at https://api.groq.com/openai/v1 (OpenAI compatibility). 200K tokens per day is roughly 100 requests of 2,000 tokens, so it suits demos and agents with short contexts.

3. OpenRouter free models

OpenRouter's pricing page lists 25+ free models from 4 free providers on its Free plan, with no platform fee. Free variants (IDs ending in :free) are capped at 20 requests per minute and 50 per day, rising to 1,000 per day once the account has bought at least $10 of credits. Its limits docs add two details: failed attempts count toward the daily cap, and a negative balance blocks free models until you top up. Browse current free models with the free price filter. If you are comparing OpenRouter with other routers, see our OpenRouter alternatives guide.

4. Cloudflare Workers AI

Workers AI is included in the Workers Free plan with 10,000 Neurons per day at no charge; paid usage beyond that is $0.011 per 1,000 Neurons, and all limits reset at 00:00 UTC (pricing). Neurons map to tokens differently per model, and Cloudflare lists models that require a paid billing method, including Kimi K2.6, GLM 5.x, and recent DeepSeek V4 Flash builds. Workers AI has an OpenAI-compatible endpoint at https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1.

5. NVIDIA NIM on build.nvidia.com

NVIDIA's catalog advertises "Free inference with leading models," featuring Nemotron models and DeepSeek V4.1 Flash on its home page (build.nvidia.com). The endpoints follow the OpenAI format at https://integrate.api.nvidia.com/v1 (API reference). NVIDIA does not publish a fixed free quota on that page, so treat it as an evaluation path, not a production dependency.

6. Vercel AI Gateway free tier

Every Vercel team gets $5 of AI Gateway credit per month on the free tier, limited to free-tier-eligible models and with lower per-model rate limits; exceeding them returns a 429 (pricing). Buying credits moves you to the paid tier, which Vercel says has zero markup on tokens.

7. Cerebras Free Trial

New Cerebras accounts receive $5 in free credits after adding a verified payment method; the credits expire 30 days after they are granted and work across public models (rate limits). It is a trial, not a permanent tier, but it is useful for testing very high-throughput inference.

8. Cohere trial keys

Cohere offers free evaluation keys limited to 1,000 API calls per month, with production keys for paid use (docs). Good for testing Cohere's embedding and rerank models in a RAG prototype.

9. Ollama (local, no API limits)

Ollama runs open-weight models on your own machine and exposes an OpenAI-compatible API at http://localhost:11434/v1. There is no rate limit and no data leaves your machine, but model size is bounded by your RAM or VRAM. If you want to run it on a rented GPU instead, see our guide to Ollama on a cloud GPU server.

10. Hugging Face Inference Providers (not free for free users)

Many lists still call Hugging Face's hosted inference free. Its pricing docs now say free users get no monthly credits and must purchase credits, while PRO users and Team or Enterprise seats get $2.00 per month. It is excellent for model variety, but budget it as paid.

The hidden costs of free LLM APIs

Hidden cost	Where it shows up	What to do
Your data trains or improves the model	Gemini free tier ("Used to improve our products: Yes")	Keep customer data on paid keys
Daily caps that include failures	OpenRouter free models	Retry with backoff, not in tight loops
Tiny tokens-per-minute	Groq Free (8K TPM on gpt-oss-120b)	Keep prompts short; batch nothing
Model availability changes	NVIDIA NIM, OpenRouter free list	Pin a fallback model
Service retirement	GitHub Models, retired July 30, 2026	Never hard-code a single free provider
Trial credits expire	Cerebras ($5, 30 days)	Use them for benchmarking, not launch

Quickstart: one client, several free providers, automatic fallback

Most free options above speak the OpenAI API format, so one client can try them in order and move on when it hits a 429. Set only the keys you have.

# pip install openai
import os
from openai import OpenAI, RateLimitError, APIStatusError

PROVIDERS = [  # (base_url, env var with key, model)
    ("https://generativelanguage.googleapis.com/v1beta/openai/", "GEMINI_API_KEY", "gemini-3.8-flash"),
    ("https://api.groq.com/openai/v1", "GROQ_API_KEY", "openai/gpt-oss-120b"),
    ("https://openrouter.ai/api/v1", "OPENROUTER_API_KEY", os.getenv("OPENROUTER_FREE_MODEL", "")),
    ("http://localhost:11434/v1", None, "llama3.2"),  # local Ollama, any model you have pulled
]

def ask(prompt: str) -> str:
    for base_url, key_env, model in PROVIDERS:
        key = os.getenv(key_env) if key_env else "ollama"
        if not key or not model:
            continue
        try:
            client = OpenAI(api_key=key, base_url=base_url, timeout=30)
            r = client.chat.completions.create(model=model, messages=[{"role": "user", "content": prompt}], max_tokens=300)
            return f"[{model}] {r.choices[0].message.content}"
        except (RateLimitError, APIStatusError) as e:
            print(f"{model} failed ({getattr(e, 'status_code', '?')}), trying next")
    raise RuntimeError("all free providers exhausted")

print(ask("Give me three names for a habit-tracking app."))

Test a single provider from the shell:

curl https://api.groq.com/openai/v1/chat/completions \
  -H "Authorization: Bearer $GROQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-oss-120b","messages":[{"role":"user","content":"Hello from the free tier"}]}'

The same fallback idea in Node.js:

import OpenAI from "openai";

const providers = [
  { baseURL: "https://generativelanguage.googleapis.com/v1beta/openai/", key: process.env.GEMINI_API_KEY, model: "gemini-3.5-flash-lite" },
  { baseURL: "https://api.groq.com/openai/v1", key: process.env.GROQ_API_KEY, model: "qwen/qwen3.8-27b" },
];

for (const p of providers.filter((p) => p.key)) {
  try {
    const client = new OpenAI({ apiKey: p.key, baseURL: p.baseURL });
    const r = await client.chat.completions.create({ model: p.model, messages: [{ role: "user", content: "One-line product tagline?" }] });
    console.log(p.model, r.choices[0].message.content);
    break;
  } catch (e) {
    console.warn(p.model, e.status ?? e.message);
  }
}

For coding agents such as Cline or Continue, our guide to the free LLM API for coding agents covers tool-specific setup.

When free runs out: the cheapest paid models

Free tiers are for building and testing. When you launch, the cheapest official list prices for text on October 21, 2026 were:

Model	Input / output per 1M tokens	Source
Gemini 2.5 Flash-Lite	$0.10 / $0.40	Google
`gpt-6-luna`	$0.10 / $0.50	OpenAI
Claude Haiku 5.5 (prompts up to 100K)	$0.10 / $0.50	Anthropic
`deepseek-flash`, off-peak	$0.15 / $0.60	DeepSeek

At these rates, 10M input and 2M output tokens a month costs about $1.80 to $2.70. Flatkey has no free tier, but it lets you reach these models with one OpenAI-compatible key and one balance; its FAQ lists bonus credit on top-ups (for example +$3 on a $10 top-up) as of October 21, 2026, and its model directory listed gpt-6-luna at $0.06 / $0.30. For a broader comparison of entry plans, see free tier showdown across major providers and AI API pricing comparison.

FAQ

Is there a completely free LLM API with no limits?

No hosted API is free without limits. Every provider above caps requests, tokens, credits, or Neurons per day or month. The only option without API limits is running an open model locally, for example with Ollama, where your hardware is the limit.

Can I use an LLM for free?

Yes. The simplest start is a Gemini API key from Google AI Studio on a free-tier model such as gemini-3.8-flash, or a Groq Free plan key for openai/gpt-oss-120b. Both work with the OpenAI SDK by changing the base URL.

What is the cheapest LLM API to use after the free tier?

On official list prices as of October 21, 2026, Gemini 2.5 Flash-Lite ($0.10 input / $0.40 output per 1M tokens), gpt-6-luna and Claude Haiku 5.5 ($0.10 / $0.50), and DeepSeek Flash off-peak ($0.15 / $0.60) are the cheapest mainstream text models.

Which free LLM API is best?

For most app prototypes, the Gemini API free tier, because it offers current Flash models, long context, and a straightforward upgrade path. Groq is best when speed on short prompts matters, and OpenRouter when you want to compare many open models.

Are free LLM APIs safe for production?

Not as your only path. Free tiers change limits without notice, some use your data to improve models, and services can be retired, as GitHub Models was on July 30, 2026. Use them for development and keep a paid fallback for launch.