Anthropic Launches Claude 3.5 Haiku, Matching Opus at Lower Cost

04-11-2024

Anthropic has unveiled Claude 3.5 Haiku, a groundbreaking AI model that matches the performance of their premium Claude 3 Opus while operating at substantially lower costs.

Written by:

Jannick

Marketing Lead

claude35haiku

Anthropic has released Claude 3.5 Haiku, a new AI model that achieves the same performance as their previous flagship Claude 3 Opus while maintaining significantly lower operational costs. The model launched on November 4, 2024, and is now available through multiple platforms including Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI.

Performance and Pricing

The new Haiku model demonstrates impressive capabilities in code generation and logical reasoning, scoring 40.6% on the SWE-bench Verified test, surpassing many publicly available models including GPT-4. Pricing is set at $1 per million input tokens and $5 per million output tokens, approximately four times higher than the previous Haiku model but still considerably more cost-effective than Opus.

Technical Specifications

Claude 3.5 Haiku features a 200,000-token context window, allowing it to process approximately 150,000 words in a single interaction. The model’s knowledge cutoff extends to July 2024, providing more recent information compared to Sonnet 3.5’s April 2024 cutoff.

Enterprise Applications

The model is particularly suited for tasks requiring rapid response times, such as customer service chatbots and real-time applications. Its improved reasoning capabilities make it effective for code suggestions, e-commerce solutions, and educational platforms. Organizations can reduce costs through prompt caching and batch processing methods.

Market Impact

This release demonstrates that high performance no longer necessarily requires high costs. Claude 3.5 Haiku’s combination of speed, cost-efficiency, and advanced capabilities positions it as a compelling option for businesses looking to implement AI solutions without significant infrastructure investments, while maintaining the high performance standards expected from enterprise-grade AI models.