Exa Folds 2 Billion Financial Documents Into One Search API

Exa has launched Exa for Finance , adding roughly 2 billion financial documents to its general-purpose search API. The index is available through the existing endpoint without a separate product or higher per-request price, according to the company.

The launch consolidates several common data sources behind one natural-language query. Agents can retrieve market prices, SEC filings, earnings transcripts, security profiles, economic releases, and financial news without maintaining a separate connector for each source.

Two billion documents, one endpoint

The index combines structured market records with the filings, transcripts, and reporting that explain them. Exa’s finance documentation lists the following coverage:

Exa previously advertised 19 million SEC filings in its broader web index. The finance documentation groups more than 2 million SEC filings and earnings transcripts together, so the published counts describe different categories. Applications with strict coverage requirements will need to verify forms, issuers, and date ranges against their own test sets.

Natural-language queries meet source filters

A request such as NVIDIA stock price and change today returns a market quote, while Tyson Foods Q4 FY2025 earnings call transcript targets the primary transcript. Developers can set category="financial report" when a query should favor filings and related financial documents.

The category filter matters for research agents because broad web searches often rank news coverage above the underlying filing or transcript. Constraining the source type gives downstream models cleaner evidence and more precise citations.

Benchmark scores need task labels

Exa claims state-of-the-art performance on FinSearchComp, a 635-question benchmark whose benchmark paper models financial research workflows. Its three tasks cover time-sensitive data retrieval, simple historical lookup, and complex historical investigation.

When Exa 3.0 launched, the company’s Exa 3.0 post reported an overall FinSearchComp score of 68.0%. Parallel, a competing search provider, later published its own head-to-head results for individual tasks:

Parallel reported that its advanced tier exceeded Exa Auto by about six percentage points on the complex investigation task at a lower cost. The 68.0% figure aggregates the benchmark, while Parallel’s T2 and T3 figures isolate tasks under a separate test setup. Product comparisons therefore depend on task mix, search tier, methodology, and query cost.

Consolidation cuts integration work

Financial applications often combine a licensed market-data feed, an SEC ingestion pipeline, a transcript provider, and a news API. Exa places much of that retrieval behind one interface, reducing connector maintenance for retrieval-augmented generation systems, where a model searches external sources before producing an answer.

Exa also offers Dynamic Highlights, which extracts query-relevant passages instead of returning full pages. The company says highlights use 95% fewer tokens than full-page text. Search-heavy agent loops can therefore send less text to the model, reducing input costs and potentially lowering latency.

Specialist providers still offer capabilities that require separate evaluation, including tick-level feeds, exchange entitlements, adjusted historical series, redistribution rights, and contractual service levels. Exa’s main economic advantage is simpler retrieval for workflows that can operate on its documented freshness and coverage.

Where one endpoint helps

The combined index is suited to workflows that need several source types during the same task:

• Equity and credit research that cross-references prices, filings, transcripts, and recent reporting
• KYC, KYB, and adverse-media screening
• Portfolio, issuer, and economic-policy monitoring
• Deal sourcing and private-market research, including funding rounds filtered by sector and date
• Risk analysis across multiple filings, issuers, or reporting periods

For multi-security research, Exa directs developers to Agent task runs. Those runs return citation-backed, schema-validated fields instead of a list of raw search results, which can simplify ingestion into databases and application interfaces.

Checks before replacing a specialist feed

Production evaluations should cover the data properties that vendor benchmarks and corpus totals cannot establish:

• Freshness: Measure quote and news latency during market hours, including timestamps and delayed-data labels.
• Historical depth: Check required tickers, SEC form types, transcript dates, delisted securities, and private companies.
• Normalization: Verify symbol mappings, corporate actions, adjusted prices, currencies, and fiscal-period conventions.
• Calculation accuracy: Reproduce derived metrics and multi-step answers from cited source records.
• Operational terms: Review rate limits, uptime commitments, data licenses, and redistribution permissions.
• Total cost: Model search requests, highlights, Agent task runs, retries, and downstream model tokens together.

Exa for Finance is available now through the existing API. Its clearest initial fit is a team already using Exa for web retrieval that wants broader financial coverage without adding several new integrations. Trading, valuation, and regulated compliance systems will require workload-specific validation before a specialist feed can be removed.