Bedrock Foundation Models Intermediate
Bedrock provides access to a diverse catalog of foundation models. Each model has different strengths, context windows, and pricing. This lesson compares the available models and provides a framework for selecting the right model for your use case.
Available Models
| Provider | Model | Context | Strengths |
|---|---|---|---|
| Anthropic | Claude 3.5 Sonnet/Haiku | 200K tokens | Reasoning, code, analysis, safety |
| Meta | Llama 3.1 (8B/70B/405B) | 128K tokens | Open weights, multilingual, code |
| Amazon | Titan Text/Embeddings | 32K tokens | Cost-effective, embeddings |
| Mistral | Mistral Large/Small | 32K tokens | Multilingual, code, efficiency |
| Cohere | Command R/R+ | 128K tokens | RAG, enterprise search |
Model Selection Framework
Decision Guide
Complex reasoning, analysis, code: Claude 3.5 Sonnet (best quality, higher cost) High-volume, cost-sensitive: Claude 3.5 Haiku or Llama 3.1 8B (fast, cheap) RAG and enterprise search: Cohere Command R+ (optimized for retrieval) Embeddings for vector search: Titan Embeddings or Cohere Embed Multilingual content: Mistral Large or Llama 3.1 Image understanding: Claude 3.5 Sonnet (vision capable)
Pricing Considerations
- On-demand pricing — Pay per input/output token. Best for variable or low-volume workloads.
- Provisioned throughput — Reserve model capacity for consistent performance. Best for high-volume production.
- Batch inference — Submit large batches at 50% discount. Best for offline processing.
Cost Strategy: Start with Claude Haiku or Llama 8B for prototyping and development. Upgrade to larger models only for use cases that demonstrate clear quality improvement. Many production workloads work well with smaller, cheaper models.
Ready to Integrate the API?
The next lesson covers how to call Bedrock models programmatically with the AWS SDK.
Next: API Integration →