Benchmarking Financial Search: How Valyu Powers Reliable AI Reasoning in Finance
Today, we’re announcing the first Search API to beat leading systems on the toughest benchmark for financial deep research.
Our API already powers millions of retrievals every day in AI applications used by financial analysts bringing machine-level accuracy and grounding to AI-powered research workflows.
Most financial search systems break under real-world conditions. They struggle to reason across noisy filings, market data streams, and time-sensitive sources with the precision analysts demand.
And the benchmarks fail too. They oversimplify workflows, relying on clean tasks, and ignore the multi-step reasoning real research requires. So we built our own: a benchmark designed to test whether a system can execute real-world analyst workflows, including parsing earnings reports, tracking macro indicators, scanning news and filings, and integrating data across sources to answer the most complex questions in finance.
The Benchmark
Our Finance Benchmark is made up of four components:
- SEC Filings: Long, noisy, and highly structured. Tests whether a system can efficiently locate relevant information in dense real-world documents.
- Financial News: Measures freshness, relevance, and the ability to retrieve time-sensitive information under pressure.
- FinLitQA: Tests financial literacy by retrieving accurate answers from finance textbooks and reference materials. Covers core concepts, definitions, and calculations.
- Multi-turn QA: Evaluates whether a system can integrate evidence across multiple queries and sources — just like an analyst.
Results: Accuracy and Relevance
Finance evaluation: Valyu reached 73% accuracy, Parallel 67%, Exa 63%, and Google 55% on financial questions.
Valyu beat incumbent systems on accuracy, multi-step reasoning, and cost-efficiency by focusing on one core idea: Build the web for machines.
Why This Matters
The benchmark results make it clear. Valyu retrieves more accurate answers from harder financial sources than any other system we tested. It handles real world workflows across filings, macroeconomic data, and textbook-level questions with higher precision and stronger grounding.
That matters because integrating financial data into AI products is slow and fragmented. The information is spread across legacy APIs, inconsistent formats, and unreliable update cycles. Most teams spend more time managing the data than building the product.
Valyu was built to change that. It is search infrastructure for agents, and AI systems that need clean, current, and trusted answers from filings, earnings calls, policy reports, and financial literature.
Try It Yourself
Don’t take our word for it — test it yourself.
Run your own retrievals on Finance GPT and see what precision at the foundation looks like.
🔑 Get your API key
📚 Read the docs
🧠 Use with LangChain