Anthropic has launched its most affordable and fastest AI model — Claude 3 Haiku, which the company claims is up to half the cost of GPT 3.5 and works up to three times faster than existing models. This newest addition to the Claude model family joins the ranks of Claude 3 Opus and Claude 3 Sonnet.
Claude 3 Haiku is a cost-effective AI solution offered by Anthropic with a fee of $0.25 per token for input and $1.25 for output. This makes it accessible to enterprises of all sizes, Anthropic wrote in a blog post.
Anthropic argued that Haiku is not only affordable but also efficient. “Businesses can rely on Haiku to quickly analyze large volumes of documents, such as quarterly filings, contracts, or legal cases, for half the cost of other models in its performance tier. For instance, Claude 3 Haiku can process and analyze 400 Supreme Court cases or 2,500 images for just one US dollar,” the blog noted.
Dario Amodei, co-founder and CEO, shared insights with VentureBeat regarding the customer segmentation for Haiku. He outlined two primary categories — latency-sensitive, and cost-sensitive. Amodei explained that latency-sensitive customers typically prioritize user-facing aspects, such as ensuring a smooth user interface. He elaborated on the significance of minimizing response times, noting that even a slight delay, such as three seconds instead of one, could lead to a loss of customers and disrupt workflow efficiency for businesses.
Claude 3 Haiku, as highlighted by AWS in a recent blog post, presents several practical applications. It excels in customer interactions, offering rapid and precise support, including translation services. Additionally, the model proves valuable in content moderation, effectively identifying and managing risky behavior or customer requests. Moreover, Claude 3 Haiku contributes to cost-saving endeavors by optimizing logistics, enhancing inventory management, and facilitating swift knowledge extraction from unstructured data.
Anthropic is offering Claude 3 Haiku through its API or a Claude Pro subscription on claude.ai. The offering is already available on Amazon Bedrock and according to Anthroopic, it will also be available on Google Cloud Vertex AI soon.
“The landscape for Generative AI models is currently experiencing a period of hyper-growth. Claude 3 is a powerful new contender in the growing large language model (LLM) market,” said Prabhu Ram, head of the Industry Intelligence Group at CyberMedia Research. “Anthropic’s competitive edge with Claude 3 rests on its focus on creating sufficient guardrails, explainability, and enduring user appeal among enterprise customers.”
Claud 3 Haiku will be three times faster
Anthropic has claimed that Haiku is three times faster than other models, processing 21,000 tokens per second for prompts under 32,000 tokens.
In addition to speed and affordability, Anthropic said it strongly emphasizes enterprise-grade security measures within Claude 3 Haiku. It said rigorous testing protocols are implemented to minimize the risk of harmful outputs and model breaches, ensuring data integrity and confidentiality.
Continuous systems monitoring, secure coding practices, and stringent access controls further bolster Haiku’s security framework, instilling confidence in enterprises entrusting their sensitive data to Anthropic’s AI solutions, the company said.
Trained on non-public data sets
Claude 3 models, including Haiku, are trained using a blend of publicly available internet data and non-public sources. Significantly, Anthropic clarifies that no training is conducted on user-generated data, irrespective of subscription status.
Beyond text comprehension, Anthropic’s Claude 3 models showcase significant progress in handling complex multimodal reasoning challenges.
As per the blog post, these models excel in tasks such as the AI2D science diagram benchmark by integrating both image and video-frame inputs. Notably, Claude 3 Sonnet leads the pack with an impressive 89.2% accuracy in the 0-shot setting, followed closely by Claude 3 Opus (88.3%) and Claude 3 Haiku (80.6%).
Copyright © 2024 IDG Communications, Inc.
