Anthropic Enhances Claude AI Models with AWS Trainium2 and Model Distillation

Posted on December 3, 2024

Anthropic has announced significant improvements to its Claude AI models through expanded collaboration with AWS. Claude 3.5 Haiku now offers latency-optimized inference on AWS Trainium2 chips, delivering up to 60% faster performance while maintaining accuracy.

The company is working with AWS on Project Rainier, an EC2 UltraCluster featuring hundreds of thousands of Trainium2 chips, promising five times more computing power than current AI model training systems.

Anthropic has also introduced model distillation capabilities in Amazon Bedrock, allowing Claude 3 Haiku to achieve accuracy levels similar to Claude 3.5 Sonnet while maintaining cost-effectiveness. This automated process includes synthetic data generation and model training without manual intervention.

Pricing updates include the Trainium2-powered Claude 3.5 Haiku at $1 per million input tokens and $5 per million output tokens, while standard Claude 3.5 Haiku prices have been reduced to $0.80 per million input tokens and $4 per million output tokens across all platforms.