Groq runs an AI inference cloud: it executes already-trained models to return responses to your applications. GroqCloud provides an API and developer tools. It is distinct from Grok, xAI’s model family.
What does Groq do?
- Runs language models through an API.
- Transcribes audio through the catalog’s Whisper models.
- Interprets images through the documented vision model.
Groq developer Console and API
The Groq Console, at console.groq.com, provides the Playground and API key creation. The API is mostly compatible with OpenAI clients, using https://api.groq.com/openai/v1 as its base URL. Check unsupported features before migrating an application.
Which models does Groq support?
As checked on October 3, 2026, production models include openai/gpt-oss-20b, openai/gpt-oss-120b, whisper-large-v3 and whisper-large-v3-turbo. Llama 3.1 8B and Llama 3.3 70B are listed for Enterprise access. Verify your tier and the current model ID in the catalog.
For vision, the docs list qwen/qwen3.8-27b, a preview model accepting text and images and returning text. Preview models are intended for evaluation and may be discontinued at short notice; they should not be presented as production models.
LPU architecture and company status
Groq developed the LPU (Language Processing Unit). It signed a non-exclusive inference technology licensing agreement with NVIDIA and continues as an independent company. In August 2026 it became an NVIDIA Cloud Partner and announced a future expansion with NVIDIA Groq 3 LPX and Vera Rubin NVL72. The announcement does not confirm general availability of that capacity on GroqCloud.
Public pricing
Developer-tier billing is based on usage. For example, openai/gpt-oss-20b costs US$0.075 per million input tokens and US$0.30 per million output tokens, checked on October 3, 2026. The Free tier has its own restrictions.
Strengths
- Console and API access for testing and integration.
- A documented catalog spanning text, transcription and vision.
Points of attention
- Check deprecations before copying an older quickstart example. Llama 3.1 8B, Llama 3.3 70B and Llama 4 Scout were retired from Free/Developer tiers, with documented Enterprise contract exceptions.
- OpenAI compatibility is partial, and model features and limits differ.
Sources
All accessed on October 3, 2026.
- Overview - GroqDocs, Groq. Accessed October 3, 2026.
- Groq Raises $650M to Scale Its AI Inference Cloud Business, Groq. Accessed October 3, 2026.
- Grok API Documentation | SpaceXAI Docs, SpaceXAI. Accessed October 3, 2026.
- Supported Models - GroqDocs, Groq. Accessed October 3, 2026.
- Images and Vision - GroqDocs, Groq. Accessed October 3, 2026.
- Quickstart - GroqDocs, Groq. Accessed October 3, 2026.
- OpenAI Compatibility - GroqDocs, Groq. Accessed October 3, 2026.
- Model Deprecation - GroqDocs, Groq. Accessed October 3, 2026.
- What is a Language Processing Unit?, Groq. Accessed October 3, 2026.
- Groq and Nvidia Enter Non-Exclusive Inference Technology Licensing Agreement to Accelerate AI Inference at Global Scale, Groq. Accessed October 3, 2026.
- Groq Among the First to Bring NVIDIA Groq 3 LPX and Vera Rubin NVL72 to Market, Groq. Accessed October 3, 2026.
- Groq Becomes an NVIDIA Cloud Partner, Groq. Accessed October 3, 2026.
- Billing FAQs - GroqDocs, Groq. Accessed October 3, 2026.
