Anthropic launches Claude 2.1 model, supporting 200K contexts and reducing illusions.

AI startup Anthropic has launched its latest conversational model, Claude 2.1, claiming to have new features that enhance enterprise applications. This version increases the context length limit of Claude to 200,000 tokens and reduces the error rate by 50%.

Some key highlights of Claude 2.1 include:

• Improved honesty, reduced illusion, and increased reliability
• Expanded context window, unlocking new use cases such as long-form content and RAG
• Early access tools and function calls for higher flexibility and scalability

Claude 2.1 represents Anthropic's ongoing efforts to balance cutting-edge AI capabilities with safety and accuracy. The updated model can now handle documents up to 150,000 words long, equivalent to over 500 pages of materials, such as technical documents, financial reports, and even literary works.

In a blog post, the company explained, "Our users can now upload entire code repositories, S-1 filings, or even lengthy literary works like 'The Iliad' or 'The Odyssey.' With the ability to process large amounts of content or data, Claude can provide summaries, answer questions, predict trends, compare and contrast multiple documents, and more."

Handling 200,000 tokens is a complex task that was previously unprecedented in the industry. Claude may only require a few minutes instead of several hours of manual work. Anthropic expects significant improvements in latency as the technology matures.

Tests have shown that Claude 2.1 has reduced the illusion or false assertion rate by half compared to the previous version, Claude 2.0. The company conducted tests on areas where AI models often make mistakes regarding factual questions, and the results showed that Claude 2.1 more frequently acknowledges uncertainty instead of providing incorrect information.

The updated model also demonstrates improved meaningful understanding and summarization capabilities, especially for long and complex documents requiring high accuracy, such as contracts, financial reports, and technical specifications. Anthropic recorded a 30% decrease in incorrect answers, and the occurrence of Claude 2.1 incorrectly supporting a claim in a document decreased by 3-4 times.

Improved developer experience and new system prompts

Anthropic has simplified the developer experience for Claude 2.1's API. The new workspace product allows for rapid iteration in a playground-like environment while offering new model settings to optimize behavior. Additionally, the introduction of system prompts enables users to set specific instructions for Claude to play a particular personality or role and provide customized responses based on user needs.

Introduction of API tool usage

Claude 2.1 also introduces a beta feature for API tool usage, allowing integration with existing systems and data sources. Early adopters can leverage Claude's language capabilities to build applications that parse natural language requests into API calls, search private databases, or perform simple operations through software. Example use cases include:

• Performing complex numerical reasoning using a calculator
• Converting requests into structured API calls
• Answering questions by searching databases or using web search APIs
• Executing simple operations in software through private APIs
• Connecting to product datasets to provide recommendations and assist users in completing purchases

The updated model is now available through Anthropic's API and powers the claude.ai website. Free users can access core functionalities, while paid users can unlock the full context window of 200,000 tokens for large-scale document analysis.