Groq API Free Tier: Free Plan Rate Limits by Model (2026)

Short answer: yes, Groq's API is free to start. Every GroqCloud organization begins on the Free plan, which gives rate-limited access to models such as OpenAI's gpt-oss-120b and gpt-oss-20b, Qwen3.8-27B, Whisper speech-to-text and Orpheus text-to-speech at no cost. Unlike many providers, Groq publishes the exact free limits per model. When you outgrow them, the Developer plan is pay-as-you-go with limits that are roughly 30 times higher for the main chat models.

Free plan limits by model

These are the Free plan values from the 'Free Plan Limits' tab on Groq's rate-limits page, checked on 17 October 2026, next to the Developer plan base limits for the same models.

Abbreviations: RPM requests per minute, RPD requests per day, TPM tokens per minute, TPD tokens per day. On the free plan, 7.2K audio seconds per hour is two hours of audio per hour, and 28.8K per day is eight hours of audio per day.

Groq calls this table 'a high level summary' and says there may be exceptions. Your organization's exact limits are on the console's Limits page.

How Groq's limits work

• Organization-level. Limits apply to the whole organization, not individual users or API keys. Extra keys do not add capacity.
• Whichever comes first. You can hit any limit type first. Groq's example: with 50 RPM and 200K TPM, sending 50 tiny requests in a minute hits the request limit even though you used almost no tokens.
• Separate input and output caps. Some organizations also have input tokens per minute (ITPM) and output tokens per minute (OTPM) limits. Hover over the TPM value on the Limits page to see an 'X in / Y out' breakdown if one applies.
• Cached tokens are free for limits. Groq states cached tokens do not count towards your rate limits.
• 429 on overflow. When you exceed a limit the API returns `429 Too Many Requests` with a `retry-after` header in seconds.

Reading the rate-limit headers

Every response includes headers you can use to throttle yourself before you hit a 429. Note that the 'requests' headers refer to the daily limit and the 'tokens' headers to the per-minute limit:

How to get a free Groq API key

1. Sign up at console.groq.com. Your account starts on the Free plan.
2. Open API Keys and click to create a key. Copy it straight away and store it as `GROQ_API_KEY`.
3. Try a model in the Playground, or call the API directly. Groq's API is OpenAI-compatible, so most OpenAI SDKs work with a changed base URL.
4. Visit Settings → Limits to see your organization's current limits.

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"}]}'

What the free plan is good for

Groq runs models on its own LPU hardware and is known for fast token generation. Within the free limits, typical fits are:

• Prototyping chat and agent apps on gpt-oss-120b. 1,000 requests and 200K tokens a day is enough for development and demos, though not for public traffic.
• Transcription side projects. Up to eight hours of audio a day through Whisper Large v3 Turbo covers podcasts, meeting notes or lecture recordings.
• Safety filtering. The Llama Prompt Guard models have the highest free daily request limit (14.4K RPD), which suits screening user input before it reaches a larger model.
• Voice prototypes with Orpheus text-to-speech, though the free limit (100 requests and 3.6K tokens a day) is small.

Note that 8K tokens per minute on gpt-oss-120b is the tightest constraint in practice. A single request with a long document can use most of a minute's budget, so keep prompts short or split work over time.

Developer plan pricing

Groq's billing FAQ describes the Developer plan as pay-as-you-go for developers and companies who want more capacity. Upgrading takes effect immediately with no upfront charge, and unlocks higher limits, chat support, Batch and Flex processing, and spend limits. Prices from Groq's models page:

Worked example: 10M input tokens and 2M output tokens on gpt-oss-120b cost 10 × $0.15 + 2 × $0.60 = $1.50 + $1.20 = $2.70. Transcribing 100 hours of audio with Whisper Large v3 Turbo costs 100 × $0.04 = $4.00.

When to upgrade: if you regularly see 429 responses on the 1,000 requests per day or 8K tokens per minute caps, or you are about to expose the app to real users, the Developer plan is the next step. At the prices above, a hobby app that uses 200K tokens a day of gpt-oss-120b (the free daily cap) would cost at most a few dollars a month on the Developer plan. Even if every token were billed at the output rate, 30 days × 0.2M tokens × $0.60 per 1M = $3.60; input-heavy traffic at $0.15 per 1M costs less.

How Groq bills

• A valid payment method is required to move from Free to Developer. Groq accepts Visa, Mastercard, American Express and Discover cards, plus US bank accounts and SEPA debit.
• Usage is invoiced monthly in arrears. New Developer accounts are also charged at progressive thresholds of $1, $10, $100, $500 and $1,000; for India billing addresses the thresholds are $1, $10 and then $100 recurring.
• Groq only bills once usage reaches at least $0.50.
• You can return to the Free plan at any time from Settings → Billing, after paying a final invoice for unbilled usage.

Ways to stretch limits and cut costs

• Prompt caching (automatic). On gpt-oss-120b, gpt-oss-20b and gpt-oss-safeguard-20b, Groq caches repeated prompt prefixes with no code changes. Cached input tokens cost half the normal input price, and because cached tokens don't count towards rate limits, a stable system prompt also stretches your free TPM. Groq says cached data expires after a period without use (the docs mention 2 hours).
• Batch API. Batch jobs cost 50% less than synchronous calls, run within a window you choose from 24 hours to 7 days, and use separate limits from your normal per-model limits. Each input file can hold up to 50,000 lines and 200MB. The batch discount does not stack with the caching discount. Groq lists Batch among the Developer plan features.
• Flex processing (paid). Adding `"service_tier": "flex"` to a request gives roughly 10 times the on-demand rate limits at the same price while capacity is available. When it is not, requests fail fast with HTTP 498 `capacity_exceeded`, so retry with jittered backoff.
• Spend limits (paid). Developer plan users can set automated spend limits and budget alerts, which is worth doing before you put a key into any public app.

Preview models and model availability

Several free-plan models, including Qwen3.8-27B, gpt-oss-safeguard-20b, the Prompt Guard models and Orpheus, are labelled preview. Groq says preview models 'are intended for evaluation purposes only and should not be used in production environments'. The production models on the list are gpt-oss-120b, gpt-oss-20b and the two Whisper models.

Groq's models page also lists llama-3.3-70b-versatile, llama-3.1-8b-instant, MiniMax M2.7 and Meta's Muse Spark with 'Contact Sales' pricing. These do not appear in the published Free plan table, so do not assume free access to them.

Data handling

According to Groq's data page, Groq does not retain customer data for inference requests by default. It may temporarily log inputs and outputs, kept for up to 30 days, when troubleshooting errors or investigating suspected abuse such as rate-limit circumvention. Any customer can turn on Zero Data Retention in Data Controls, which also disables features that need storage such as batch jobs. Customer data that is retained is stored in Google Cloud buckets in the United States.

Groq vs other free AI APIs

• Gemini API free tier: includes proprietary Gemini models, but limits are not published and free-tier data may be used by Google.
• Cerebras inference free trial: another fast-inference provider serving open models.
• OpenRouter free models: routes to free endpoints across many providers.
• Cloudflare Workers AI: a free daily allowance tied to a platform you may already use. (GitHub Models was retired on July 30, 2026.)

For a full side-by-side list, see our free AI API keys guide.