Amazon Bedrock – Consistent Anthropic FM pricing across regions

Mihaly Balassy - Jul 4 - - Dev Community

AWS typically offers varying prices for each service across its global regions. However, if we look at Amazon Bedrock Anthropic On-Demand and Batch prices we see a different pattern. They are consistent across regions. The primary variation lies in the model versions available in each region. As new regions are continuously added to Amazon Bedrock, it’s a good idea to look at what each region offers.

Amazon Bedrock is available in 13 AWS regions:

  • US East (N. Virginia)
  • US West (Oregon)
  • Asia Pacific (Tokyo)
  • Asia Pacific (Singapore – limited access)
  • Asia Pacific (Sydney)
  • Asia Pacific (Mumbai)
  • Canada (Central)
  • Europe (London)
  • Europe (Frankfurt)
  • Europe (Paris)
  • Europe (Ireland – limited access)
  • South America (São Paulo)
  • AWS GovCloud (US-West)

💡 Note: Due to limited access in Ireland, Singapore and GovCloud, these regions are excluded from my analysis, leaving us with 10 regions for comparison.

Available Anthropic models

Anthropic uses "Haiku" for its smallest model, "Sonnet" for the mid-range option, and "Opus" for its top-tier model.

On June 21, 2024, Anthropic introduced Claude 3.5 Sonnet, claiming it can match or exceed OpenAI's GPT-4o or Google's Gemini across numerous tasks. This new model is exclusively available through the Anthropic API, Amazon Bedrock, and Google Cloud's Vertex AI. Since this is a new model, it is currently only available from us-east-1 at the time of this blog post.

When it comes to Amazon Bedrock, the rule of thumb you've come to know about AWS pricing and service availability no longer applies. If you are based in Europe, you have learned that the region with the most services is Ireland (eu-west-1) and that the cheapest option is usually Stockholm (eu-north-1).
With Bedrock, this all changes; the region that offer the most Anthropic FM's for Europe is Frankfurt (eu-central-1). If you're developing with Generative AI in Europe, Frankfurt is now your best choice.
In the US, your rule of thumb remains the same, with us-east-1 being the region that provides the most functionality, but there is a catch: the highest-end model, Claude 3 Opus is only available in Oregon.

AWS Anthropic pricing

Amazon Bedrock's on-demand and batch pricing is consistent across regions (e.g., you pay the same for Claude 3 Haiku in Oregon and Canada). In fact, Claude Sonnet 3 and 3.5 cost the same. This uniform pricing strategy underscores AWS's commitment to making advanced Generative AI models accessible and affordable for developers.

Provisioned Throughput pricing

When we look at the provisioned throughput pricing, we see fewer choices. We are limited to Claude Instant and Claude 2.0/2.1 models and with only 4 regions. Frankfurt is right behind the top US regions in terms of prices, with Tokyo being the most expensive.

Anthropic Claude Instant Provisioned Throughput pricing

Anthropic Claude 2.0/2.1 Provisioned Throughput pricing

Summary

AWS demonstrates its strong commitment to Generative AI through multiple actions. It offers competitive pricing for Anthropic's foundation models, continually expands Bedrock's availability to new regions, and promptly makes the latest foundation models accessible to its users.

These efforts highlight AWS's dedication to advancing and supporting Generative AI, providing developers with the latest tools at affordable prices.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .