Groq

Groq is an LLM inference API, sold as GroqCloud, that serves models from OpenAI, Meta and Qwen to developers, billed per token with a rate-limited free tier.

Groq runs trained AI models for applications that need answers quickly, so a team can call a model over an API instead of hosting it. Developers sign up for GroqCloud at console.groq.com and send requests to models served from Groq's own data centers. Groq LLC, founded in 2016, owns and operates the service, and it remains an independent company after licensing its inference technology to NVIDIA in December 2025. The model list covers chat and reasoning models such as OpenAI's GPT-OSS, a Qwen model that also reads images, Whisper for speech to text and Orpheus for text to speech. Tool use lets a model call the developer's own functions, and built-in server-side tools and MCP extend that further. Structured Outputs return responses that conform to a supplied JSON schema. A reasoning_effort setting trades speed for deeper thinking on the models that support it. Prompt caching cuts the cost of repeated prefixes, the Batch API takes offline jobs at a lower rate, and Data Controls let any customer switch on Zero Data Retention. GroqCloud suits developers and product teams building chat, agents or voice features where response time matters. A rate-limited free tier lets a developer try each model before paying anything. Adding a payment method moves the account to the Developer plan, which charges per token for each model with no plan fee and unlocks higher limits, batch jobs and Flex processing. Companies that need the Performance tier, LoRA adapter serving or the Llama models priced on request agree an Enterprise plan with Groq's sales team. Typical work is real-time chat and multi-step agent loops, transcription of calls and recordings with Whisper, and large offline jobs sent through the Batch API. Teams whose code already calls models through the OpenAI client libraries can point it at GroqCloud, and Groq also ships its own Python and JavaScript SDKs. A Prometheus integration lets operators watch inference traffic beside the rest of their production monitoring. Prompt Guard and GPT-OSS Safeguard models screen user input before it reaches the main model in a chat or agent pipeline.