How to get a free Groq API key

Groq (with a q) runs open models such as OpenAI's gpt-oss and Qwen on its own LPU chips and is known for very fast answers, often several hundred tokens per second. Its API has a free plan that needs no payment method, which makes a Groq key one of the easiest free keys to use in BYOK apps. Not to be confused with Grok, xAI's model: that has its own Grok API key guide.

Step by step

1. Sign in at console.groq.com, or create a free account. New accounts start on the Free tier.
2. Open API Keys at console.groq.com/keys and create a key. Name it after the app you'll use it in.
3. Copy the key right away and paste it into the app's settings.
4. Pick a model the free plan includes, for example openai/gpt-oss-120b. If the app has no Groq option, add a custom OpenAI-compatible provider with base URL https://api.groq.com/openai/v1 and your Groq key.

Free plan limits

Groq's rate limits page listed these Free plan limits on 2026-10-29:

Model	Requests / min	Requests / day	Tokens / min	Tokens / day
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 and -turbo (speech to text)	20	2,000	7,200 audio seconds / hour	28,800 audio seconds / day
canopylabs/orpheus-v1-english (text to speech)	10	100	1,200	3,600

What that means in practice: a chat turn with some history is often about 2,000 tokens in and 500 out, so 200,000 tokens a day is roughly 80 turns, and the 8,000 tokens per minute cap allows about three such turns a minute. Long conversations and pasted documents reach the token limits long before the request limits. Whisper's 28,800 seconds is eight hours of audio a day, which is plenty for dictation apps.

Limits apply to your whole organization, not per key, and any one of them can trigger. When you hit one, the API returns error 429 with a retry-after header, and most apps simply wait and retry. Cached tokens don't count toward the limits.

Llama is no longer on the free plan. Groq's models page lists Llama 3.3 70B and Llama 3.1 8B as Enterprise models. If an app defaults to llama-3.3-70b-versatile and fails, switch it to gpt-oss or Qwen.

The paid Developer tier

Add a payment method under Settings → Billing to upgrade. There is no upfront charge: you are billed for tokens at the end of the month, or earlier when your spend crosses thresholds ($1, $10, $100 and up). Limits jump to 1,000 requests per minute and 250,000 tokens per minute for gpt-oss, and you can set spend limits. Prices per million tokens on 2026-10-29:

Model	Input	Output
openai/gpt-oss-120b	$0.15	$0.60
openai/gpt-oss-20b	$0.075	$0.30
qwen/qwen3.8-27b	$0.80	$4.00
whisper-large-v3-turbo	$0.04 per hour of audio

A typical chat turn on gpt-oss-120b costs about $0.0006, so $5 lasts for thousands of messages. You can downgrade back to the Free tier from the Billing settings at any time.

Privacy

Groq says it does not keep the inputs and outputs of inference requests by default. It may log them temporarily (up to 30 days) when troubleshooting failures or investigating abuse, and features such as batch jobs and fine-tuning store files until you delete them. The Data Controls settings in the console let you turn those features off.

Keep the key safe

• One key per app, so you can revoke a single key if an app misbehaves.
• Never put the key in a website's front-end code or a public repository.
• On the Developer tier, set a spend limit before giving the key to a new app.

When something else fits better

Groq's free plan covers mid-sized open models. For Google's own models for free, use the Gemini free tier; for a wider choice of free open models behind one key, see OpenRouter's free models; for full privacy and no limits at all, run a model on your own computer with Ollama or LM Studio. All the options are compared in free AI API keys.