Claude 2.1

Anthropic has released a new model, Claude 2.1, through the API and claude.ai. This update includes an expanded context window, reduced hallucination rates, and tool use capabilities to enhance the performance of enterprise AI applications.

The biggest change is the introduction of a 200K context window. This is enough to handle massive amounts of data at once—roughly 150,000 words or over 500 pages. Users can upload entire codebases, financial statements, or lengthy literary works to perform summarization, Q&A, and trend forecasting.

The model's reliability has also been significantly improved. Claude 2.1 achieves a 2x reduction in hallucination rate compared to the previous model, Claude 2.0. In particular, its ability to acknowledge uncertainty rather than provide incorrect information on complex factual questions has improved, increasing enterprise trust.

Key improvements include:

• Improved comprehension and summarization: Reduced incorrect answer rates by 30% on complex documents such as legal documents and technical specifications, and lowered the rate of misinterpreting document content by 3-4x.
• Tool use: A new beta feature that enhances the model's extensibility.
• Cost efficiency: Updated pricing policies for each model to reduce cost burdens for customers.