State-of-the-art across research benchmarks.

DRACO evaluates long-form deep research systems on accuracy against cost. It measures how well agents produce grounded research outputs across leading deep-research products.

Valyu DeepResearch leads at 72.7% accuracy at $2,500 per 1k requests — higher accuracy than Perplexity, Claude, Gemini and OpenAI deep research, at a lower cost.

FreshQA is updated roughly every week and contains 600 time-sensitive questions that assess how effectively APIs handle recent and evolving information. The benchmark is critical for use cases such as news summarization, event tracking, and trend monitoring. Our evaluation used the Aug 27th test set, the most recent release available at the time of testing.

FreshQA was evaluated by comparing four search solutions (Valyu, Parallel, Google, Exa) on time-sensitive queries. Valyu achieved the highest performance at 79.0% accuracy, significantly outperforming others.

SimpleQA, created by OpenAI, includes 4,326 factual questions designed to measure retrieval precision on straightforward, unambiguous queries. The benchmark serves as the baseline for evaluating general-purpose search quality and factual accuracy, covering diverse topics from science and technology to TV shows and video games.

Valyu achieved 94% accuracy on SimpleQA, outperforming other search APIs.

The Economics Benchmark evaluates search performance through 100 of the hardest analyst-style questions curated from authoritative economic data sources. The benchmark draws from World Bank Indicators, Bureau of Labor Statistics (BLS), and Federal Reserve Economic Data (FRED), covering GDP, inflation, employment, productivity, and monetary policy to simulate authentic economic research workflows used in policy analysis and macroeconomic forecasting.

The 100 questions are sourced from: World Bank Indicators (50 questions), BLS (25 questions), and FRED (25 questions), representing the most challenging queries from each dataset.

Valyu achieved 72.6% overall accuracy, substantially outperforming Parallel (52%), Exa (44.7%), and Google (42.7%) on complex economic queries.

The Financial Benchmark assesses API performance through 120 queries distributed across seven datasets covering critical financial domains. The benchmark simulates real-world financial information retrieval scenarios across regulatory, academic, and market-specific contexts.

The 120 queries span seven financial datasets: SEC Filings, Papers, Financial news and reports, Stocks, Forex, Insider Transactions and Crypto.

Valyu achieved 72.5% overall accuracy, significantly outperforming Exa (62.5%) and Google (55%) on financial queries.

MedAgent is a collection of the most difficult questions curated from 10 leading medical benchmarks. The benchmark comprises 482 challenging questions spanning clinical knowledge, drug information, and medical reasoning, designed to test the limits of medical AI capabilities. It includes specialized datasets from Clinical Trials and Drug Labels alongside questions from established medical benchmarks including MMLU, MEDQA, and others.

The 482 questions are sourced from: MMLU (50), MEDQA (50), MMLU-PRO (50), MEDEXQA (50), PUBMEDQA (50), MEDMCQA (50), MEDBULLETS (50), AFRIMEDQA (32), MEDXPERTQA-U (50), MEDXPERTQA-R (50).

MedAgent was evaluated by comparing four different search solutions (Valyu, Google, Exa, Parallel) with Valyu achieving the highest performance of 48.1% overall accuracy with particularly strong results on Clinical Trials and Drug Labels (87.5% each).