What Is Claude 3? Anthropic’s New AI Models Explained

A complete guide to Anthropic's newest AI developments: Claude 3 Opus, Sonnet, and Haiku.

The race among artificial intelligence (AI) startups is heating up more than ever. Anthropic, a company founded by alumni of OpenAI, has launched its latest and most sophisticated generative AI (Gen AI) model: Claude 3.

Claude 3 represents the cutting-edge in Anthropic's large language model (LLM) offerings, taking center stage in powering the Claude AI chatbot. This suite encompasses three distinct models: the top-tier Claude 3 Opus, the mid-range Claude 3 Sonnet, and the yet-to-be-released Claude 3 Haiku.

Anthropic asserts that Claude 3 surpasses its previous version, Claude 2.1, and competes favorably against industry giants like OpenAI's GPT-4 and Google's Gemini Ultra. But what makes Claude 3 stand out among its counterparts like ChatGPT and GPT-4? Here's what we've discovered.

Exploring Claude 3: Opus, Sonnet, and Haiku

Claude 3 represents the forefront of Anthropic's LLM advancements, backed by significant investment, including a $4 billion infusion from Amazon. The company is committed to developing AI technologies responsibly, aligning with the effective altruism movement to benefit society broadly, not just its shareholders.

This new suite of LLMs, named Claude 3, comprises three unique models:

• Claude 3 Opus: Anthropic's top-tier model, providing unparalleled intelligence and sophistication.
• Claude 3 Sonnet: A versatile model that combines performance with speed, suited for many users.
• Claude 3 Haiku (remains unreleased): Focused on efficiency and speed, perfect for quick data processing and customer interactions.

Both Claude 3 Opus and Sonnet are now accessible through claude.ai and the Claude API, and they are available in 159 countries. Haiku is slated for release soon.

Google Cloud customers will soon have access to all three models via API in the Vertex AI Model Garden, with immediate opportunities to apply for private preview access to Claude 3 Sonnet.

Claude 3 Pricing Overview

Claude 3 offers different models with prices for both personal and business use:

• Claude 3 Opus: For personal use, it costs $20 a month. Businesses pay $15 for every million input tokens and $75 for every million output tokens.
• Claude 3 Sonnet: Available for free for personal use. Businesses pay $3 for every million input tokens and $15 for every million output tokens.
• Claude 3 Haiku: Pricing is based on usage, with businesses paying $0.25 for every million input tokens and $1.25 for every million output tokens.

For more details on pricing, you can visit their website.

Advanced Vision Features

Claude 3 models offer powerful vision features and can handle photos, charts, and diagrams. This is great for businesses with lots of visual data, like PDFs and slides, making it easier to work with a large part of their information.

Fast and Efficient Responses with Claude 3

Claude 3 models boost the speed of real-time interactions, such as live chat support, quick text completions, and fast data extraction.

• Haiku is the quickest and most cost-effective. It analyzes complex documents in less than three seconds, and its speed will improve further after launch.
• Sonnet is twice as fast as earlier versions, Claude 2 and 2.1, and smarter, too. It is great for fast information retrieval and sales support.
• Opus has the same speed as its predecessors but offers greatly enhanced intelligence.

Claude 3 vs GPT-4: Preliminary Insights

We're in the early stages of evaluating Claude 3 against leading competitors like GPT-4 from ChatGPT. Currently, we haven't finalized our analysis to share comprehensive insights.

However, Anthropic has published several performance evaluations of Claude 3, claiming it competes well with major models like ChatGPT's GPT-4 and Google's Gemini Ultra LLM. It's important to note that these assessments are from Anthropic itself and may favor scenarios that highlight Claude 3's strengths.

This doesn't mean Claude 3 lacks capability; early indications suggest it's a formidable model. Yet, a conclusive comparison with GPT-4, Gemini, or other AI solutions awaits our thorough review.

Reduced Refusal Rates

Previous Claude versions could unnecessarily refuse prompts, indicating a possible lack of understanding. The Claude 3 series, including Opus, Sonnet, and Haiku, shows significant progress, with fewer inappropriate refusals. These models now better understand queries near their limits without overreacting, accurately identifying actual risks while reducing unnecessary denials of harmless requests.

Increased Accuracy

The accuracy of these models is essential for businesses that rely on them. To assess this, a series of challenging, fact-oriented questions are used, targeting the models' known weaknesses. Responses are categorized into correct answers, incorrect answers (also known as hallucinations), and instances where the model admits uncertainty instead of providing inaccurate information. Compared to Claude 2.1, Opus demonstrates a significant improvement, offering twice the accuracy on these difficult questions and reducing the frequency of errors.

Committed to Responsible Design

Anthropic has dedicated teams focused on addressing a wide range of risks, from misinformation and child sexual abuse material (CSAM) to biological threats, election interference, and the potential for autonomous replication. Anthropic continues to advance safety and transparency initiatives, such as Constitutional AI and has refined the models to address privacy concerns introduced by new features.

Although the Claude 3 models have improved in areas such as biological and cyber knowledge and autonomy, they are currently classified as AI Safety Level 2 (ASL-2) under Anthropic's Responsible Scaling Policy. Red teaming evaluations, conducted in accordance with commitments made to the White House and the 2023 US Executive Order, have determined that the models currently pose a negligible catastrophic risk.

Getting Started with Claude 3

Interested in testing out Claude 3's capabilities? You're in luck—Claude 3 Sonnet is now available on Claude AI across numerous countries worldwide.

Here's a simple guide to get started:

1. Visit the Anthropic website.
2. Select the "Try Claude" option.
3. Sign in with your Google account (or create a new account if you don't have one).

And that's all there is to it! You'll be taken directly to the main Claude AI chat interface, where you can begin exploring what Anthropic's assistant offers. If you're impressed by its performance, you might then consider upgrading to Opus for even more features.

Further Reading:

AAnthropic'sannouncement

The Claude 3 Model Family: Opus, Sonnet, Haiku