A new Anthropic API feature that helps some Claude models link responses directly to source documents could significantly reduce errors and avoid hallucinations, the company said.
The new feature, Citations, is intended to improve document summarization, Q&A and customer support applications, Anthropic said in a blog post.
Citations allows Anthropic AI models to provide detailed references to “exact sentences and passages” from documents used to create responses.
Citations is only available on Claude 3.5 Sonnet and Claude 3.5 Haiku via the Anthropic API and Google Cloud’s Vertex AI platform.
Sourcing a 100-page document as a reference would cost approximately 30 cents with Claude 3.5 Sonnet, or 8 cents with Claude 3.5 Haiku, based on Anthropic’s API pricing.
The new API is significant, Anthropic said, since it now lets user add source documents to the context window, and when querying the model, Claude automatically cites claims in its output that are inferred from those sources.
Previously, developers relied on complex prompts that instructed Claude to include source information, which often led to “inconsistent performance and significant time investment in prompt engineering and testing,” according to the company.
Anthropic pointed to two use test cases of Citations in its blog post. It said Thomson Reuters is deploying Claude to power its CoCounsel legal AI reference platform to help legal and tax professionals synthesize expert knowledge.
“It makes citing and linking to primary sources much easier to build, maintain, and deploy to our users. This capability not only helps minimize hallucination risk but also strengthens trust in AI-generated content,” Jake Heller, head of product, CoCounsel, Thomson Reuters, said in a statement.
Financial technology company Endex, meanwhile, is using Claude to operate an autonomous agent for financial firms.
“With Anthropic’s Citations, we reduced source hallucinations and formatting issues from 10% to 0% and saw a 20% increase in references per response. This removed the need for elaborate prompt engineering around references and improved our accuracy when conducting complex, multi-stage financial research,” Endex CEO Tarun Amasa said in a statement, also on Anthropic’s blog.
