Anthropic’s new citations feature aims to reduce AI errors

In an announcement that may have been timed to shift attention away from OpenAI Operatorunveiled on Anthropic Thursday A new feature of its developer API is called Citationswhich allows developers to “subtract” answers from Claude’s family from Amnesty International In source documents such as emails.

Anthropic says Citations allows its AI models to provide detailed references to “exact sentences and passages” from the documents it uses to generate responses. As of Thursday afternoon, citations were available in both Anthropic’s API and Google Vertex AI platform.

As the Anthropist explains in A Blog post Using citations, developers can add source files to have models automatically cite the claims they infer from those files. Citations are especially useful in document summarization, Q&A, and customer support applications, Anthropic says, as this feature can prompt forms to include source citations.

Citations are not available for all Anthropic models—only the Claude 3.5 Sonnet and Claude 3.5 Haiku. Also, the feature is not free. Anthropic notes that citations may charge a fee depending on the length and number of source documents.

Based on Anthropic’s standard API pricing, which Citations uses, a 100-page source document will cost approximately $0.30 with Claude 3.5 Sonnet, or $0.08 with Claude 3.5 Haiku. This may be worth it for developers looking to scale back Hallucinations And other errors caused by artificial intelligence.