Hi, I’m Dang, an AI engineer at Knowledgelabo, Inc. We provide a service called "Manageboard", which supports our clients in aggregating, analyzing, and managing scattered internal business data. Manageboard is set to enhance its AI capabilities in the future.
This article shares some of the challenges we faced during R&D and offers implementation tips—particularly on how to choose the right model available in AWS Bedrock.
Background
AWS Bedrock is a managed service that lets you access multiple foundation models through a unified interface. In this article, we’ll compare and organize Bedrock-compatible models from the following perspectives:
- Types and characteristics of models (Claude, Titan, Nova)
- Availability in the AWS Tokyo Region
Supported Models
As of July 2025, here are the main serverless models available in the Tokyo region (ap-northeast-1):
| Model | Provider | Primary Use Case | Features |
|---|---|---|---|
| Claude (Sonnet / Haiku) | Anthropic | Text generation / RAG | High accuracy, strong safety features, excellent long-context handling |
| Titan / Nova | Amazon | Embedding / Text generation | Bedrock-native models, low latency, fast and lightweight |
Claude (by Anthropic)
Anthropic’s Claude series is known for its high-accuracy, safety-oriented LLMs.
| Model | Key Features | Input Cost (USD / 1K tokens) | Output Cost (USD / 1K tokens) |
|---|---|---|---|
| Sonnet | Balanced middle-tier model | 0.003 | 0.015 |
| Haiku | Fast and lightweight | 0.00025 | 0.00125 |
- Claude Opus is not available in Tokyo
- Haiku is suitable for lightweight Japanese tasks
- Well-suited for knowledge-based RAG applications
Titan / Nova (Amazon)
Amazon provides its own LLMs under Titan and Nova.
Titan Models
| Model | Purpose | Input Cost (USD / 1K tokens) | Output Cost (USD / 1K tokens) |
|---|---|---|---|
| Titan Text Embeddings | Text vectorization | 0.0002 (V1) / 0.000029 (V2) | — |
| Titan Text Express | Text generation | 0.000275 | 0.000825 |
- Lightweight and cost-effective
- Titan Text Embeddings is used for converting documents into vectors when building knowledge bases
- Titan Text Express is often used for quick prototyping in the Playground
Nova Models
| Model | Input Cost (USD / 1K tokens) | Output Cost (USD / 1K tokens) |
|---|---|---|
| Nova Micro | 0.000042 | 0.0000105 |
| Nova Lite | 0.000072 | 0.000018 |
| Nova Pro | 0.00096 | 0.00024 |
- Nova Premier is not available in Tokyo
- Suitable for chatbots and agents
- Fast response
- Note: While Claude has higher output token costs, Nova has higher input token costs
Limitations in Tokyo Region
As of July 2025, the following AI-related features and services are not available in AWS Tokyo Region (ap-northeast-1):
| Feature | Availability Note |
|---|---|
| Claude Opus / Nova Premier | Not available in Tokyo Region |
| Custom model import | Not available in Tokyo Region. Unstable Japanese support elsewhere |
| Image input | Not yet supported by LLM blocks or Agents |
| AWS Textract (OCR service) | Not available in Tokyo Region. Lacks Japanese support elsewhere |
Summary
AWS Bedrock offers excellent integration with AWS infrastructure and allows flexible model selection depending on use case.
Here are our practical recommendations:
- Claude Sonnet / Haiku: RAG, FAQs, or anything requiring high precision
- Titan: Document embedding for search or knowledge base
- Nova: For general-purpose chat or agents where speed is critical
Discussion