# Groq (LLM inference) — card for AI agents

Fast inference API for open models such as GPT-OSS and Llama, priced per 1M tokens.

Maker: Groq · category: Model hosting and inference

## Price (official page)
- Free tier: none listed
- Usage price: gpt-oss-120b $0.15 in / $0.60 out per 1M tokens; gpt-oss-20b $0.075 / $0.30
- Pricing page: https://console.groq.com/docs/models (checked 2026-10-09)

## Connect
- npm: `npm install groq-sdk`
- PyPI: `pip install groq`
- Repository: https://github.com/groq/groq-typescript

## For AI agents on Groq's own site (AgentGid check 2026-11-13: 2 of 6)
- llms.txt: none found
- OpenAPI description: none found at the usual places
- Official MCP server: no · official SDK: yes · status page: no
- robots.txt lets AI assistants (ChatGPT-User, Claude-User) read the site: yes

## Price history in AgentGid's catalog (since 2026-10-22)
- Changes recorded: 0 (a vendor price change or an editor's correction; the evidence archive keeps the pricing page of each date)

## Other APIs in this category
- OpenAI API: https://agentgid.com/apis/openai-api/agent.md
- Gemini API: https://agentgid.com/apis/google-gemini-api/agent.md
- Mistral AI API: https://agentgid.com/apis/mistral-api/agent.md
- Cohere: https://agentgid.com/apis/cohere/agent.md
- fal: https://agentgid.com/apis/fal/agent.md

## Links
- AgentGid page: https://agentgid.com/apis/groq/


## Groq: fast answers from open AI models

Groq is an online service that runs open AI language models, such as OpenAI's GPT-OSS, and returns their answers very quickly. Your app sends a question or a task and gets the text back, paying per amount of text. It works with the same code as the OpenAI service once you change the address and the key.

### Where to use it
- A chat assistant in your app: Answer your users' questions inside your app or site, with replies that show up word by word while they are written.
- Text features in a Next.js app: Write product descriptions, summaries or recipes from a short prompt inside a JavaScript web app built with the Vercel AI toolkit.
- Sort and tag incoming emails: Turn free text, like support emails or form answers, into fixed fields your app can store, such as category and urgency.
- Answers with today's news: Let the model look things up on the web before it answers, for questions about recent events.
- Voice notes into text: Turn recorded voice messages or calls into text with OpenAI's Whisper model, hosted by Groq and billed per hour of audio.
- Big jobs overnight at half price: Send thousands of requests at once, such as tagging a whole product catalog, and collect the results within a day or a week.

### How to start
- Create an API key in the Groq console and keep it in your environment as GROQ_API_KEY.

```
export GROQ_API_KEY=<your-api-key-here>
```

- Install the official Python package (or keep using the OpenAI package with Groq's address).

```
pip install groq
```

- To reuse OpenAI code, change only the base address and the key.

```
client = openai.OpenAI(
    base_url="https://api.groq.com/openai/v1",
    api_key=os.environ.get("GROQ_API_KEY")
)
```


### License
Groq is a paid online service with a free tier, and you can use it in a commercial product. You own your inputs and outputs; each model also has its own maker's terms you must follow.

### Good to know
- You need a Groq account and a secret key. The free tier needs no card; paying per use needs a payment method.
- Limits count for your whole organization, so all your keys and apps share them.
- Some popular models, such as Llama 3.3 70B, are now listed for enterprise customers only, with prices on request.
- Preview models can disappear at short notice, so use production models for a live app.
- Using several accounts to get around the limits, spam, and fully automated high-risk decisions about people are not allowed.

Checked 2026-11-10 against: https://groq.com, https://groq.com/pricing, https://console.groq.com/llms.txt, https://console.groq.com/docs/overview.md, https://console.groq.com/docs/quickstart.md, https://console.groq.com/docs/openai.md, https://console.groq.com/docs/models.md, https://console.groq.com/docs/model/openai/gpt-oss-120b.md, https://console.groq.com/docs/model/llama-3.3-70b-versatile.md, https://console.groq.com/docs/rate-limits.md, https://console.groq.com/docs/errors.md, https://console.groq.com/docs/api-reference.md, https://console.groq.com/docs/text-chat.md, https://console.groq.com/docs/structured-outputs.md, https://console.groq.com/docs/tool-use/overview.md, https://console.groq.com/docs/tool-use/built-in-tools/browser-search.md, https://console.groq.com/docs/speech-to-text.md, https://console.groq.com/docs/batch.md, https://console.groq.com/docs/libraries.md, https://console.groq.com/docs/billing-faqs.md, https://console.groq.com/docs/spend-limits.md, https://console.groq.com/docs/your-data.md, https://console.groq.com/docs/production-readiness/security-onboarding.md, https://console.groq.com/docs/legal/ai-policy.md, https://console.groq.com/docs/legal/services-agreement.md, https://console.groq.com/docs/legal/customer-data-processing-addendum.md, https://github.com/groq/groq-typescript/blob/main/README.md, https://registry.npmjs.org/groq-sdk/1.6.0, https://pypi.org/pypi/groq/1.7.0/json

### Mistakes to avoid
- Don't: Copying model llama-3.3-70b-versatile from the quickstart into a new app. → Do: Use a production model listed with a price, such as openai/gpt-oss-20b or openai/gpt-oss-120b. (The models page now lists Llama 3.3 70B as Enterprise with Contact Sales pricing.)
- Don't: Setting dangerouslyAllowBrowser to true and calling Groq from the web page. → Do: Call Groq from a server route and keep GROQ_API_KEY on the server. (Anyone who opens the page could copy the key and spend your money.)
- Don't: Sending OpenAI options logprobs, logit_bias or n greater than 1. → Do: Leave them out; Groq answers them with a 400 error. (These OpenAI fields are not supported on Groq.)
- Don't: Asking for strict JSON output and web search in the same request. → Do: Run browser search first, then a second request with structured outputs on the answer. (Browser search does not work together with structured outputs.)
- Don't: Using whisper-large-v3-turbo to translate audio into English. → Do: Use whisper-large-v3 for translation; turbo only transcribes. (The speech-to-text table marks translation as not supported for turbo.)

### Questions
- Is there a free tier? Yes. The Free tier needs no payment method; you add one only to move to the pay-as-you-go Developer tier.
- Can I keep my OpenAI code? Mostly. Point the OpenAI client at https://api.groq.com/openai/v1 with a Groq key; a few options like logprobs are not supported.
- Does Groq train on my prompts? No. Its agreement says inputs and outputs are not used for training unless you allow it.
- Can I cap my bill? Yes, on paid plans. A spend limit blocks all keys of the organization when the monthly cap is reached.
- Am I charged for failed requests? Not for errors on Groq's side (500, 502, 503).

---
AgentGid's catalog is curated: every entry was checked against the maker's official pages (registry, repository, license and pricing pages) on the date shown; a license value shows the quote it stands on wherever AgentGid has one. Usage numbers come from AgentGid's own scan of public repositories set up for AI coding agents. Claims by makers are reported, not tested by AgentGid.
