Get Started →
AWS Foundation Models

AWS Bedrock Pricing Calculator

Estimate costs for foundation models from Anthropic (Claude), Meta (Llama), Amazon (Nova), Mistral AI, Cohere, OpenAI, and more. Calculate pricing for text generation, embeddings, images, video, and optional Bedrock features like Guardrails, Knowledge Bases, and Flows.

100+ Models 16 Providers US East Region 2026 Pricing
⚠️
Pricing Accuracy Notice

This calculator uses pricing data based on US East (N. Virginia) region. Pricing may vary by AWS region and change over time. For the most accurate and up-to-date pricing, please visit the official AWS Bedrock Pricing page. Always verify costs before production deployment.

Configure Your Bedrock Usage

Pricing shown is for US East (N. Virginia). Other regions may have different rates.
Average number of input tokens per month (in 1000s)
Average number of output tokens per month (in 1000s)

Optional Bedrock Features

▼
Content filtering, denied topics, sensitive information redaction ($0.15-$0.17 per 1K text units)
Vector database for RAG applications ($5/GB storage + $1 per 1K retrievals)
Multi-step AI workflows ($0.035 per 1K node transitions)

Advanced Pricing Options

▼
Save up to 90% on cached tokens (75% discount on cache writes, 90% on cache reads)
Available for select models - non-time-sensitive workloads
Automatically route to optimal models ($1.00 per 1K requests)
Automatically improve prompts ($0.03 per 1K tokens)

About AWS Bedrock Pricing

Amazon Bedrock offers access to high-performing foundation models from leading AI providers including Anthropic (Claude), Meta (Llama), Amazon (Nova, Titan), Mistral AI, Cohere, AI21 Labs, Stability AI, and more. Pricing varies by model provider, model size, usage type, and AWS region.

📋 Official Pricing Source: All pricing data in this calculator is based on the AWS Bedrock Pricing page for the US East (N. Virginia) region. Regional pricing variations apply. Always verify current rates on the official AWS website before deployment.

How Bedrock Pricing Works

AWS Bedrock uses consumption-based pricing with different models for various use cases:

  • Text Models: Charged per 1,000 tokens for input and output separately. Larger models with more parameters typically cost more.
  • Image Generation: Charged per image based on resolution and quality settings.
  • Embeddings: Charged per 1,000 tokens processed for vector generation.
  • Batch Inference: 50% discount available for select models when using batch processing.
  • Regional Variations: Pricing differs by AWS region. US East typically has the lowest rates.

Popular Foundation Models (US East Region Pricing)

Prices shown are for US East (N. Virginia) and may vary in other regions. Verify on AWS official page

  • Claude Sonnet 4.6 (Anthropic): $3.00 per million input tokens, $15.00 per million output tokens. Latest balanced model with superior performance (2026).
  • Llama 4 Scout (Meta): $0.27 per million input tokens, $0.36 per million output tokens. Efficient model with 10M context window.
  • Amazon Nova Pro: $0.80 per million input tokens, $3.20 per million output tokens. Excellent price-performance for general tasks.
  • Mistral Large 3: $4.00 per million input tokens, $12.00 per million output tokens. Latest flagship for enterprise use.
  • GPT-4o (OpenAI): $2.50 per million input tokens, $10.00 per million output tokens. Latest multimodal GPT model.

Optional Bedrock Features

Guardrails: Apply content filtering, denied topics, sensitive information redaction, and automated reasoning checks. Pricing ranges from $0.10 to $0.17 per 1,000 text units depending on the guardrail type.

Knowledge Bases: Build RAG applications with managed vector storage. Costs $5 per GB per month for storage plus $1.00 per 1,000 standard retrievals.

Flows: Create multi-step AI workflows with visual builder. Charged at $0.035 per 1,000 node transitions.

Prompt Routing: Automatically route requests to optimal models at $1.00 per 1,000 requests.

Cost Optimization Tips

  • Use prompt caching for repeated context to save up to 90% on cached tokens
  • Choose smaller models (like Nova Lite or Claude Haiku) for simpler tasks
  • Leverage batch inference for 50% discount on non-time-sensitive workloads
  • Use Intelligent Prompt Routing to automatically select cost-effective models
  • Implement efficient prompting techniques to reduce token usage

Model Comparison by Use Case

Best for Complex Reasoning: Claude Opus 4.6 ($5/$25), Claude Sonnet 4.6 ($3/$15), o1 ($15/$60), Llama 3.1 405B ($5.32/$16)

Best Price-Performance: Amazon Nova Pro ($0.80/$3.20), Claude Haiku 4.5 ($1/$5), Llama 3.1 70B ($2.65/$3.50), GPT-4o Mini ($0.15/$0.60)

Most Cost-Effective: Amazon Nova Micro ($0.035/$0.14), Nova Lite ($0.06/$0.24), Llama 3.1 8B ($0.30/$0.60), Llama 4 Scout ($0.27/$0.36)

Best for Code: Claude Sonnet 4.6 ($3/$15), Llama 3.1 70B ($2.65/$3.50), o1-mini ($3/$12), Qwen3 Coder 30B ($0.65/$0.65)

Best for Long Context: Claude models (200K tokens), Llama 4 Scout (10M tokens), Llama 4 Maverick (1M tokens), Kimi K2 Thinking (long-context)

Best for Embeddings: Titan Embeddings v2 ($0.02/M), Cohere Embed v3 ($0.10/M)

Best for Image Generation: Stable Image Core ($0.030/image), Nova Canvas ($0.040/image), SDXL 1.0 ($0.040/image)

Service Tiers Explained

  • Standard (On-Demand): Pay-as-you-go pricing with no commitments. Best for variable workloads and development/testing.
  • Batch Inference: 50% discount on standard pricing for non-time-sensitive workloads. Results delivered within 24 hours.
  • Provisioned Throughput: Reserved capacity with guaranteed throughput. Requires commitment (1-month or 6-month). Best for predictable production workloads.
  • Priority Tier: Premium access with lower latency and higher throughput limits. Higher pricing than standard.
  • Flex Tier: Discounted pricing for flexible, non-urgent workloads with variable latency.

Regional Pricing Differences

Bedrock pricing varies by AWS region. US East (N. Virginia) typically offers the lowest prices as AWS's primary region. Key points about regional pricing:

  • US East (N. Virginia): Base pricing region - usually the lowest rates
  • US West (Oregon): Typically within 0-5% of US East pricing
  • Europe Regions: Often 5-15% higher than US East due to data center costs
  • Asia Pacific: May be 10-20% higher depending on specific region
  • Model Availability: Not all models are available in all regions - check AWS documentation
  • Verify Before Deployment: Always check the official AWS Bedrock pricing page for your specific region

⚠️ Important: This calculator shows US East (N. Virginia) pricing. For production cost estimates in other regions, use the region selector on the official AWS Bedrock pricing page to see exact rates for your deployment region.

Understanding Token-Based Pricing

Most text models charge separately for input and output tokens. A token is roughly 4 characters or 0.75 words in English. For example:

  • 100 words ≈ 133 tokens
  • 1,000 words ≈ 1,333 tokens
  • A typical page of text ≈ 500-750 tokens

Output tokens typically cost 3-5x more than input tokens because generation is more computationally intensive than processing.

Real-World Pricing Examples

Example 1: Customer Support Chatbot

Scenario: 10,000 customer queries per month, average 200 tokens input, 300 tokens output per query.

Model: Claude Haiku 4.5 ($1.00/$5.00 per million tokens)

  • Total input tokens: 2 million (10,000 × 200)
  • Total output tokens: 3 million (10,000 × 300)
  • Input cost: 2M × $1.00/M = $2.00
  • Output cost: 3M × $5.00/M = $15.00
  • Monthly Total: $17.00

Example 2: Document Summarization Pipeline

Scenario: Process 1,000 documents/month, average 5,000 tokens input, 500 tokens output per document. Using batch inference for 50% discount.

Model: Llama 3.1 70B ($2.65/$3.50 per million tokens, batch: $1.33/$1.75)

  • Total input tokens: 5 million (1,000 × 5,000)
  • Total output tokens: 500,000 (1,000 × 500)
  • Input cost: 5M × $1.33/M = $6.65
  • Output cost: 0.5M × $1.75/M = $0.88
  • Monthly Total: $7.53 (vs $15.05 without batch discount)

Example 3: RAG Application with Knowledge Bases

Scenario: 50GB knowledge base, 50,000 queries/month with retrieval, average 100 tokens input + 500 context tokens, 400 tokens output.

Model: Claude Haiku 4.5 ($1.00/$5.00 per million tokens) + Knowledge Bases

  • KB Storage: 50GB × $5/GB = $250.00
  • KB Retrievals: 50K × $1.00/1K = $50.00
  • Total input: 30M tokens (50K × 600)
  • Total output: 20M tokens (50K × 400)
  • Model input cost: 30M × $1.00/M = $30.00
  • Model output cost: 20M × $5.00/M = $100.00
  • Monthly Total: $430.00

Example 4: Image Generation Service

Scenario: Generate 5,000 images per month for marketing content.

Model: Stable Image Core ($0.030 per image)

  • 5,000 images × $0.030 = $150.00
  • Monthly Total: $150.00

Frequently Asked Questions

Is this calculator accurate for my AWS region? +
This calculator uses pricing from US East (N. Virginia), which is AWS's base pricing region. Other regions may have different rates (typically 5-20% higher). For accurate pricing in your specific region, visit the official AWS Bedrock pricing page and select your region from the dropdown menu.
Where can I verify the exact pricing for my region? +
Visit the official AWS Bedrock Pricing page at https://aws.amazon.com/bedrock/pricing/ and use the region selector to view exact pricing for your deployment region. Always verify pricing before production deployment to avoid unexpected costs.
How much does AWS Bedrock cost? +
AWS Bedrock pricing varies by model and provider. Text models typically range from $0.15 to $15 per million tokens. For example, Claude 3.5 Sonnet costs $3.00 per million input tokens and $15.00 per million output tokens, while Amazon Nova Lite costs $0.06 per million input tokens and $0.24 per million output tokens. Image generation ranges from $0.004 to $0.08 per image depending on resolution and model.
What is the difference between Bedrock pricing tiers? +
Bedrock offers multiple pricing tiers: Standard (pay-as-you-go on-demand), Priority (premium tier with lower latency and higher throughput), Flex (discounted pricing for non-urgent workloads), Reserved (committed capacity with discounts), and Batch (50% savings for batch processing). This calculator shows Standard tier pricing, which is the most common option.
How much do Claude models cost on Bedrock? +
Anthropic Claude pricing on Bedrock varies by model: Claude 3.5 Sonnet costs $3/$15 per million tokens (input/output), Claude 3 Opus costs $15/$75, Claude 3 Sonnet costs $3/$15, and Claude 3 Haiku costs $0.25/$1.25 per million tokens. Claude models support up to 200K context windows with prompt caching available for cost savings.
What are AWS Bedrock Guardrails costs? +
Bedrock Guardrails pricing depends on the policy type: Content Filters and Denied Topics cost $0.15 per 1,000 text units, Sensitive Information Filters cost $0.10 per 1,000 text units, and Automated Reasoning checks cost $0.17 per 1,000 text units. Word filters are free. A text unit equals approximately 300 characters.
How much does Bedrock Knowledge Bases cost? +
Bedrock Knowledge Bases costs $5 per GB per month for index storage, $1.00 per 1,000 standard retrieval requests, and $4.00 per 1,000 agentic retrieval API calls. Multimodal document parsing, embeddings generation, and re-ranking are included at no extra charge when using managed services.
Does Bedrock pricing vary by AWS region? +
Yes, Bedrock pricing can vary slightly by AWS region. US East (N. Virginia) typically has the lowest prices, while regions like Asia Pacific may have slightly higher costs. Model availability also varies by region. Not all models are available in all regions, so check the AWS Bedrock documentation for region-specific availability.
What is Bedrock batch inference pricing? +
Bedrock batch inference offers a 50% discount compared to on-demand pricing for select foundation models. Batch processing is ideal for non-time-sensitive workloads like content summarization, data classification, or bulk analysis. Supported models include Claude, Llama, Mistral, and select Amazon models.
How can I reduce my Bedrock costs? +
To reduce Bedrock costs: (1) Use prompt caching to save up to 90% on repeated context, (2) Choose smaller models like Nova Lite or Claude Haiku for simpler tasks, (3) Leverage batch inference for 50% savings, (4) Use Intelligent Prompt Routing to automatically select cost-effective models, (5) Implement efficient prompting to reduce token usage, (6) Use Reserved capacity for predictable workloads, (7) Monitor usage with CloudWatch to identify optimization opportunities.
What is prompt caching and how does it work? +
Prompt caching stores frequently-used context (like system prompts or long documents) to avoid reprocessing. Cache writes cost 75% less than regular input tokens, and cache reads cost 90% less. This is ideal for RAG applications, multi-turn conversations, or any scenario with repeated context. Caches are maintained for 5 minutes and automatically refreshed with use.
Which is cheaper: Claude or Llama models? +
Llama models are generally more cost-effective with symmetric pricing (input = output). Llama 3.1 70B costs $0.99 per million tokens for both input and output, while Claude 3.5 Sonnet costs $3/$15. However, Claude models often produce better results with fewer tokens, potentially offsetting the higher per-token cost. For budget-conscious projects, consider Llama 3.1 8B ($0.22/$0.22) or Amazon Nova Lite ($0.06/$0.24).
Does Bedrock charge for embeddings? +
Yes, embedding models charge per token processed. Amazon Titan Embeddings v2 costs $0.02 per million tokens (most affordable), while Cohere Embed models cost $0.10 per million tokens. When using Knowledge Bases, embeddings generation is included at no extra charge if using managed embedding models.
Are there free tiers or credits for Bedrock? +
AWS offers free tier credits for new accounts (up to $200 for various services including Bedrock). There is no ongoing free tier for Bedrock - all usage is billed. However, you can start with small experiments using affordable models like Nova Micro ($0.035/$0.14 per million tokens) to minimize costs during development.
Can I use multiple models in one application? +
Yes! Use different models for different tasks to optimize costs. For example: Claude 3.5 Sonnet for complex analysis ($3/$15), Nova Pro for general queries ($0.80/$3.20), Nova Lite for simple tasks ($0.06/$0.24), and Titan Embeddings for vector search ($0.02/M). Intelligent Prompt Routing can automate model selection based on query complexity.

Data Source: Pricing data is based on AWS Bedrock Official Pricing for US East (N. Virginia) region as of January 2026.
Disclaimer: Estimates are provided for informational purposes only. Actual costs may vary based on region, usage patterns, and AWS pricing changes. Always verify current pricing on the official AWS website before making production decisions.
Not Affiliated: This calculator is an independent tool and is not officially affiliated with or endorsed by Amazon Web Services.

Found this tool helpful?

Share it with your team!

Twitter Facebook LinkedIn