Loading page
Groq
Fast LLM inference on custom LPU hardware
Inference specialist🇺🇸 USinferencellmfast
Last reviewed Mar 14, 2026
Groq runs LLM inference on its custom LPU (Language Processing Unit) hardware, offering an OpenAI-compatible API for open-weight models alongside dedicated bare-metal capacity for production inference.
LLM Models
7
From / 1M input
$0.08
LLM API Pricing
Pay-per-token pricing. Prices shown per 1M tokens.
Prices last updated: September 9, 2026
Pricing
| Model | Input/1M | Output/1M |
|---|---|---|
| $0.040/sec | - | |
| $0.075 | $0.300 | |
| $0.150 | $0.600 | |
| $0.600 | $3.00 | |
| $0.800 | $4.00 |
Groq pricing by model
Input, output, and batch rates, plus alternatives, for one model at a time.
Pros & Cons
Advantages
- High published token speeds (500 t/s on GPT-OSS 120B, 1,000 t/s on GPT-OSS 20B)
- OpenAI-compatible API for easy integration
- Competitive pricing for open-weight models
- Free tier available for testing
- Data centers on four continents, including the US, Canada, Finland, the UK, Australia, and Saudi Arabia
Limitations
- Small model catalog: production text models are limited to OpenAI GPT-OSS
- Focus on inference only - no training capabilities
- Several models are preview-only and may be discontinued at short notice
- Enterprise-tier models require contacting sales rather than self-serve pricing
Key Features
LPU-Powered Inference
Custom Language Processing Units run inference, with 256 LPUs and 128 GB of on-chip SRAM per rack
OpenAI-Compatible API
Drop-in replacement for the OpenAI API at api.groq.com/openai/v1, including a Responses API
Layered Platform
GroqMetal bare-metal infrastructure, GroqCore inference stack, and GroqAssured enterprise governance, each including the layers below
Multimodal Endpoints
Text generation plus speech-to-text, text-to-speech, OCR and image recognition, and content moderation models
Agentic Systems and Built-In Tools
Compound systems that combine models with web search, code execution, remote MCP tools, and Google Workspace connectors
Service Tiers
Performance, flex, and batch processing tiers, plus prompt caching, structured outputs, and LoRA inference
Free Tier Available
Rate-limited free access for experimentation and small projects
Pricing Options
| Option | Details |
|---|---|
| Pay-per-token | Token-based pricing with separate input/output rates; audio models are billed per hour and speech models per character |
| Batch and flex processing | Alternative service tiers for throughput-tolerant or bursty workloads instead of the standard on-demand tier |
| Enterprise capacity | Dedicated capacity and enterprise-only models quoted through sales |
| Free tier | Rate-limited free access for development and testing |
Availability & Support
Regions
13 data centers across four continents, including sites in the US (Liberty Lake, St Paul, Dallas, Houston), Canada (Kamloops, Vaudreuil-Dorion, Calgary), Finland, the UK, Australia, and Saudi Arabia
Support
Documentation and API reference, developer community, production readiness guides, and enterprise onboarding via sales
Getting Started
- 1
Create an account
Sign up at console.groq.com with email or OAuth
- 2
Get API key
Generate an API key from the console dashboard
- 3
Make API calls
Use the OpenAI-compatible endpoint with your preferred model
Frequently Asked Questions
Which models does Groq serve?
Check the pricing table above for the models Groq currently serves and their per-token rates.
How do I get started with Groq?
Create an account, Get API key, Make API calls
What are Groq's main advantages?
Groq's main advantages include: High published token speeds (500 t/s on GPT-OSS 120B, 1,000 t/s on GPT-OSS 20B), OpenAI-compatible API for easy integration, Competitive pricing for open-weight models, Free tier available for testing, Data centers on four continents, including the US, Canada, Finland, the UK, Australia, and Saudi Arabia.
What are Groq's limitations?
Groq's main limitations include: Small model catalog: production text models are limited to OpenAI GPT-OSS, Focus on inference only - no training capabilities, Several models are preview-only and may be discontinued at short notice, Enterprise-tier models require contacting sales rather than self-serve pricing.
Compare Providers
Find the best prices for the same GPUs and models from other providers
All Groq Comparisons
Groq vs Deep InfraGroq vs IO.NETGroq vs Novita AIGroq vs OpenRouterGroq vs Together AIGroq vs Amazon AWSGroq vs CerebrasGroq vs DigitalOceanGroq vs GMI CloudGroq vs ScalewayGroq vs EcoHashGroq vs HyperstackGroq vs LyceumGroq vs AnthropicGroq vs CohereGroq vs CrusoeGroq vs fal.aiGroq vs Fireworks AIGroq vs Google CloudGroq vs Microsoft AzureGroq vs Mistral AIGroq vs OpenAIGroq vs PerplexityGroq vs ReplicateGroq vs RuncrateGroq vs RunpodGroq vs Wafer