Anthropic's new Claude 3 Haiku can process 400 Supreme Court cases in a single US dollar

Anthropic has launched Claude 3 Haiku, an AI model that offers impressive speed, affordability, and security features. The model is designed for enterprise users who require fast and cost-effective analysis of large datasets, making it an ideal solution for many applications.

With its unmatched speed, Claude 3 Haiku can process 21K tokens (~30 pages) per second for prompts under 32K tokens, making it three times faster than its competitors. It is also capable of generating swift output, which enables responsive chat experiences and the simultaneous execution of multiple tasks.

Anthropic’s pricing model for Claude 3 Haiku is specially made for enterprise workloads, offering an input-to-output token ratio of 1:5. This cost-effective approach will let businesses to quickly analyze large volumes of documents, such as quarterly filings, contracts, or legal cases, at a much cheaper price compare to other similar LLMs. For example, Claude 3 Haiku can process and analyze 400 Supreme Court cases or 2,500 images for just one US dollar.

Here is a summary of the table:

• Claude 3 performs best on most tasks, followed by you (Gemini 1.0) and GPT-3.5.
• All three models perform well on tasks such as “Knowledge Q&A” and “Common Knowledge,” which involve accessing and processing information from the real world.
• All three models perform worse on tasks such as “Graduate Level Reasoning” and “Math Problem Solving,” which require more complex reasoning and problem-solving skills.

In addition to its speed and affordability, Claude 3 Haiku also uses enterprise-grade security. Anthropic conducts testing to reduce the likelihood of harmful outputs and jailbreaks of the mode.

Starting today, customers can access Claude 3 Haiku through the Anthropic API or with a Claude Pro subscription on claude.ai. The model is also available on Amazon Bedrock, with plans to make it available on Google Cloud Vertex AI.

More here.