Systems
Groq offers two compound AI systems that intelligently use external tools to provide more accurate, up-to-date, and capable responses than traditional LLMs alone. Both systems support web search and code execution, but differ in their approach to tool usage.
- Compound Beta (
compound-beta) - Full-featured system with up to 10 tool calls per request
- Compound Beta Mini (
compound-beta-mini) - Streamlined system with up to 1 tool call and average 3x lower latency
| Feature | Compound Beta | Compound Beta Mini |
|---|
| Tool Calls per Request | Up to 10 | Up to 1 |
| Average Latency | Standard | 3x Lower |
| Token Speed | ~350 tps | ~350 tps |
| Best For | Complex multi-step tasks | Quick single-step queries |
- Multiple Tool Calls: Can perform up to 10 server-side tool calls before returning an answer
- Complex Workflows: Ideal for tasks requiring multiple searches, code executions, or iterative problem-solving
- Comprehensive Analysis: Can gather information from multiple sources and perform multi-step reasoning
- Use Cases: Research tasks, complex data analysis, multi-part coding challenges
- Single Tool Call: Performs up to 1 server-side tool call before returning an answer
- Fast Response: Average 3x lower latency compared to Compound Beta
- Direct Answers: Perfect for straightforward queries that need one piece of current information
- Use Cases: Quick fact-checking, single calculations, simple web searches
Both systems support the same set of tools:
- Web Search - Access real-time information from the web
- Code Execution - Execute Python code automatically
- You need comprehensive research across multiple sources
- Your task requires iterative problem-solving
- You're building complex analytical workflows
- You need multi-step code generation and testing
- You need quick answers to straightforward questions
- Latency is a critical factor for your application
- You're building real-time applications
- Your queries typically require only one tool call
Both systems use the same API interface - simply change the model parameter to compound-beta or compound-beta-mini to get started.