The partnership between Amazon Web Services and Anthropic has reached new heights, with Anthropic designating AWS as its primary training partner. This expanded collaboration introduces Project Rainier, an immense AI compute cluster powered by AWS Trainium 2 chips, set to revolutionize the training and deployment of Anthropic’s advanced foundation models. Users of Amazon Bedrock will immediately benefit from a latency-optimized Claude 3.5 Haiku, offering up to 60% faster inference, and a pioneering model distillation feature for Claude 3 Haiku, delivering Claude 3.5 Sonnet-like accuracy at unmatched cost-effectiveness.
The landscape of artificial intelligence is evolving at an unprecedented pace, driven by strategic alliances that push the boundaries of what’s possible. A prime example is the deepening collaboration between Amazon Web Services (AWS) and Anthropic, a partnership that began in September of last year with Anthropic naming AWS its primary cloud provider and a significant investment from Amazon. This alliance has now intensified, with Anthropic officially designating AWS as its primary training partner, in addition to its ongoing role as the main cloud provider.
This commitment underscores Anthropic’s reliance on AWS’s robust infrastructure, specifically its cutting-edge Trainium and Inferentia chips, for training and deploying its future large foundation models. This strategic alignment is designed to continuously enhance the already premium performance, security, and privacy that Amazon Bedrock offers to customers utilizing Claude models.
As Matt Garman, AWS CEO, aptly put it, “The response from AWS customers who are developing generative AI applications powered by Anthropic in Amazon Bedrock has been remarkable.” He further emphasized that this collaboration will keep “pushing the boundaries of what customers can achieve with generative AI technologies.” Dario Amodei, co-founder and CEO of Anthropic, echoed this sentiment, stating his anticipation for “working with Amazon to train and power our most advanced AI models using AWS Trainium, and helping to unlock the full potential of their technology.”
Project Rainier: The Hyperscale Engine for Next-Generation AI
At the heart of this deepened collaboration lies Project Rainier, Amazon’s ambitious initiative to build an expansive AI compute cluster. This project, which commenced last year, is distributed across multiple data centers within the U.S. It incorporates an astounding nearly half-a-million of Amazon’s proprietary Trainium 2 chips, designed specifically for high-performance AI training.
Project Rainier is not just about scale; it’s about sheer processing power. This EC2 ultracluster of TRN2 ultraservers is projected to deliver more than five times the computing power (in exaflops) used to train the current generation of leading AI models. Anthropic’s commitment to this infrastructure is profound, with plans to scale its usage to over one million Trainium 2 chips across AWS by the end of this year. This immense power will be instrumental in training and deploying Anthropic’s largest foundation models and future iterations of Claude.
Unleashing Speed and Efficiency: Claude 3.5 Haiku on Trainium 2
A significant outcome of this collaboration is the optimization of Claude 3.5 Haiku to run on AWS Trainium 2. This optimization brings latency-optimized inference to Amazon Bedrock, making the model remarkably faster without compromising its renowned accuracy. For developers and businesses, this means tangible performance improvements.
By enabling latency optimization, Claude 3.5 Haiku can deliver up to 60% faster inference speed. This makes it the ideal choice for demanding use cases that require near-instantaneous responses, such as real-time content moderation, dynamic chatbots, and efficient code completions. This faster version, powered by Trainium 2, is currently available in public preview in the US East (Ohio) region via cross-region inference. Initially, it is offered at a competitive price of $1 per million input tokens and $5 per million output tokens.
Democratizing Frontier AI: Model Distillation in Amazon Bedrock
Beyond raw speed, the partnership is also pioneering new methods for accessibility and cost-effectiveness with the introduction of model distillation in Amazon Bedrock. This innovative technique is being applied to Claude 3 Haiku, Anthropic’s most cost-effective model from the previous generation.
Model distillation enables Claude 3 Haiku to achieve significant performance gains, reaching an accuracy level comparable to Claude 3.5 Sonnet for specific tasks, all while maintaining the original price and speed of Claude 3 Haiku. This process involves transferring knowledge from a “teacher” model (like Claude 3.5 Sonnet) to a “student” model (Claude 3 Haiku), allowing sophisticated tasks such as retrieval augmented generation (RAG) and complex data analysis to be performed at a fraction of the cost.
Unlike traditional fine-tuning, which often demands manual effort in crafting training examples and adjusting parameters, Amazon Bedrock model distillation automates the entire workflow. This includes generating synthetic training data from the teacher model, training and evaluating the student model, and hosting the final distilled model for inference. The system automatically applies various data synthesis methods, from generating similar prompts to creating high-quality responses based on user-provided examples. This powerful feature for Claude 3 Haiku in Amazon Bedrock is now available in preview.
For more details on Anthropic’s Claude 3 family and their capabilities, interested users can consult the official Anthropic news blog.
Enhanced Value: New Pricing for Claude 3.5 Haiku Across All Platforms
In a move to make Claude 3.5 Haiku even more accessible across a broader spectrum of use cases, Anthropic is also implementing a price reduction. Regardless of whether customers access the model through the Anthropic API, Amazon Bedrock, or Google Cloud’s Vertex AI, the pricing has been lowered to $0.80 per million input tokens and $4 per million output tokens. This significant adjustment reflects a commitment to democratizing advanced AI, making it more feasible for a wider range of applications and budgets.
The Broader Impact: Amazon Bedrock’s Growing Ecosystem and Responsible AI
The continued collaboration highlights Amazon Bedrock’s pivotal role as the go-to platform for building and scaling generative AI applications. Since its general availability in September 2023, over 10,000 customers have leveraged Bedrock, with many utilizing Anthropic’s Claude models. These customers, ranging from startups to enterprises like ADP, Broadridge, and Intuit, are driving innovation across various sectors.
The Claude 3 model family, including Opus, Sonnet, and Haiku, brought advanced capabilities such as vision (understanding images, charts, diagrams), best-in-class benchmarks, and significantly reduced hallucination through constitutional AI techniques. These features, combined with the ease of access and secure customization options like Retrieval Augmented Generation (RAG) on Amazon Bedrock, empower developers to create transformative experiences.
Customers like Lonely Planet, LexisNexis Legal & Professional, and Broadridge have shared testimonials on how Claude on Amazon Bedrock has revolutionized their operations, from personalizing travel itineraries and enhancing legal research to automating regulatory reporting. The commitment to responsible AI development, through frameworks like constitutional AI and adherence to industry voluntary commitments, ensures that this powerful technology is deployed ethically and safely.
The future promises an unprecedented level of innovation, with generative AI automating complex processes, augmenting human expertise, and reshaping digital experiences. The deepened collaboration between AWS and Anthropic, through advancements like Project Rainier, Trainium 2, and model distillation, provides the essential tools for enterprises and innovators worldwide to confidently navigate this next frontier of generative AI-powered innovation. To get started with Claude in Amazon Bedrock, visit the official AWS Bedrock Claude page.
