Supported LLM Models
This page details the supported Large Language Models (LLMs), their compatibility with each agent type, and recommendations for optimal performance.
Supported Models
The full list of supported models and their current status can be retrieved from GET /v1/models. Use the filter[agentType] query parameter to filter models by agent type (e.g., ?filter[agentType]=tool).
Models are added and updated over time. Use GET /v1/models for the most current list of available models, their context windows, and supported agent types.
Agent Type Compatibility
Not all models support all agent types. Tool and Task agents require models with function calling and tool choice capabilities. RAG agents can use any supported model.
| Agent Type | Requirement |
|---|---|
| RAG | All supported models |
| Tool | Models with function calling and tool choice support |
| Task | Models with function calling and tool choice support |
Recommended Models
| Use Case | Recommended Model | Model ID |
|---|---|---|
| Tool orchestration & complex reasoning | Claude 3.7 Sonnet | anthropic.claude-3-7-sonnet-20250219-v1:0 |
| General-purpose (RAG, Tool, Task) | Claude 3.5 Sonnet | anthropic.claude-3-5-sonnet-20240620-v1:0 |
| Fast & cost-effective RAG | Amazon Nova Micro | amazon.nova-micro-v1:0 |
| Fast tool workflows | Claude 3.5 Haiku | anthropic.claude-3-5-haiku-20241022-v1:0 |
| Higher-accuracy RAG | Amazon Nova Lite | amazon.nova-lite-v1:0 |
Choosing a Model
By Use Case
- High-accuracy RAG with large documents: Choose models with large context windows to fit more retrieved content.
- Tool orchestration and complex reasoning: Claude models excel at understanding and executing multi-step tool operations.
- Cost-effective simple tasks: Amazon Nova Micro and Claude 3.5 Haiku provide the best balance of speed and cost for straightforward operations.
- Structured data processing (Task Agents): Claude models offer strong instruction adherence for template-based workflows. Nova Micro is a cost-effective alternative for well-defined schemas.
By Priority
| Priority | Recommended Approach |
|---|---|
| Lowest latency | Amazon Nova Micro, Claude 3.5 Haiku |
| Strongest reasoning | Claude 3.7 Sonnet, Claude 3.5 Sonnet |
| Largest context | Check GET /v1/models for context window sizes |
| Best cost/performance | Amazon Nova Micro, Claude 3.5 Haiku |
Model Deprecation
Some models will be deprecated over time as new models are supported. Deprecation will be scheduled in advance. The effective date and the recommended model to migrate to will be specified in the response from the GET /v1/models endpoint.
Agent Builder will continue to allow agents that use models scheduled for deprecation to be created, updated, and invoked until the deprecation date. At the deprecation date, requests that use the deprecated model will fail with an error response that includes the deprecation status and replacement model ID. No further actions will be permitted with this agent until the model has been updated.