How to Get a Free Groq API Key

Groq runs open-weight models on its own inference hardware and serves them through an API that is mostly compatible with OpenAI's client libraries. Every GroqCloud organization starts on the Free plan: you create a key in the console and call the hosted models within per-model rate limits. A payment method is only needed if you upgrade to the paid Developer plan.

What you get

• Rate-limited access to the models in Groq's Free plan table. As of October 2026 that covers the text models openai/gpt-oss-120b, openai/gpt-oss-20b and qwen/qwen3.8-27b, the speech-to-text models whisper-large-v3 and whisper-large-v3-turbo, the Orpheus text-to-speech models, and the moderation models openai/gpt-oss-safeguard-20b and Llama Prompt Guard 2.
• Free-plan limits as of October 2026: 30 requests per minute (RPM), 1,000 requests per day (RPD), 8,000 tokens per minute (TPM) and 200,000 tokens per day (TPD) on the GPT-OSS 120B, GPT-OSS 20B and Qwen 3.8 27B models. Whisper allows 20 RPM, 2,000 RPD and 7,200 audio seconds per hour (28,800 per day).
• An OpenAI-compatible endpoint at https://api.groq.com/openai/v1, plus Groq's own Python and TypeScript SDKs.
• Cached prompt tokens don't count toward your rate limits.
• No payment method needed unless you upgrade to the paid Developer plan.

Lineups and limits change, so check the live table on Groq's rate limits page. Full details are on the Groq free tier page.

Steps

1. Go to console.groq.com and sign up or log in. Your account starts on the Free plan.
2. Open the API Keys page (console.groq.com/keys) and create a key. Copy it somewhere safe.
3. Save the key as an environment variable, for example export GROQ_API_KEY=your_key in ~/.zshrc or ~/.bashrc. Groq's SDKs read GROQ_API_KEY.
4. Install an SDK: pip install groq for Python or npm install --save groq-sdk for JavaScript. If your code already uses the OpenAI SDK, keep it and set the base URL to https://api.groq.com/openai/v1 with your Groq key.
5. Send a test chat completion to a model from the Free plan table, such as openai/gpt-oss-20b or openai/gpt-oss-120b. Some examples in Groq's docs still use llama-3.3-70b-versatile; Groq's deprecation page lists that model and llama-3.1-8b-instant as shut down for free and Developer usage since August 16, 2026.
6. Check your organization's exact limits under Settings > Limits (console.groq.com/settings/limits) before you build anything that depends on volume.

Never commit the key to a repository or ship it in client-side code.

Limits and the catch

• Low per-minute token caps. 8,000 TPM on the main text models is enough for testing and light apps, but long prompts or parallel requests hit it quickly. Any limit can trigger first: 30 small requests in a minute block you even if you are far under the token cap.
• Limits are per organization, not per user or per key. Extra keys in the same organization share the same quota.
• Not every model has the same limits. As of October 2026 openai/gpt-oss-safeguard-20b is capped at 3 RPM and 2,000 TPM, and the Orpheus TTS models at 10 RPM and 100 RPD. Groq notes there may be exceptions; your Limits page is authoritative.
• Preview models can disappear. Groq labels qwen/qwen3.8-27b and several other models as preview, meant for evaluation and liable to be discontinued at short notice. Production models get a deprecation notice and a replacement, but they are retired too.
• Data handling. Groq says it does not retain inference inputs and outputs by default, but may log them for up to 30 days to troubleshoot reliability problems or investigate abuse. Any customer can turn on Zero Data Retention under Settings > Data Controls.
• Upgrading costs money. The Developer plan raises limits and adds Batch and Flex processing, billed pay-as-you-go. It requires a credit card, U.S. bank account or SEPA debit account.

Other free LLM APIs

If Groq's limits or model list don't fit, two other providers offer a free way to call models:

• Mistral: Mistral Studio starts in Free mode with API access enabled and no card required. Usage and rate limits apply per model and are shown on the Limits page of the Mistral Admin Panel. The Free plan has no monthly API credits. See Mistral AI free plan.
• Cerebras: adding a payment method gets a one-time $5 credit that expires 30 days after activation. You are not charged or moved to paid usage automatically, and access pauses when the credit runs out. Free-trial limits are 5 RPM, 30,000 uncached TPM and 1 million tokens per day per model. See Cerebras Inference free trial.

For Google's models, see how to get a free Gemini API key. For a side-by-side view, see the free LLM API comparison.

If it doesn't work

• 429 Too Many Requests: you hit an RPM, RPD, TPM or TPD limit. The retry-after header says how many seconds to wait; add retries with backoff. The x-ratelimit-remaining-requests header tracks your daily requests and x-ratelimit-remaining-tokens your tokens per minute.
• 401 Unauthorized: the key is missing or invalid. Check that GROQ_API_KEY is set in the shell running your code, or create a new key.
• 403 Forbidden on a model: an admin may have restricted models for your organization or project under model permissions.
• Requests to an old model ID fail: the model has been retired. Check Groq's deprecations page for the recommended replacement.
• 400 from OpenAI-style code: Groq rejects logprobs, logit_bias, top_logprobs and messages[].name, and n must be 1. Remove those fields.