Cloud Partnerships

Parallel partners with Fiscal.ai to access financial data. When AI agents are asked a financial question, like a company's adjusted EBITDA or the exact filing date of an annual report, accuracy alone isn't enough. The answer needs to be traceable back to a real source. Parallel builds web infrastructure for AI agents, with APIs including Search, Extract, Task, FindAll, Responses, and Monitor, powered by a proprietary index of the open web. To turn financial research that used to take people days into agentic workflows that take seconds, Parallel needed data that was not just accurate, but citable down to the underlying filing. The challenge: financial data the open web doesn't expose cleanly. Parallel believes the open web is the best source, but for gaps, Parallel works with specialized providers like Fiscal.ai. Structured financial data is one such gap: fundamental figures, filing dates, and other financial fields are often buried in investor-relations pages or SEC filings that aren't built for programmatic retrieval, and for an AI agent to be trustworthy, it can't just return a number; it needs to point back to exactly where that number came from. Why Parallel chose Fiscal.ai. Fiscal.ai checked the boxes that mattered most: * Source-level citations: every response includes a URL, and where relevant, the exact SEC filing and page number * Coverage of financial fields the open web doesn't expose cleanly * A responsive team that got Parallel set up quickly "Our experience working with the Fiscal.ai API has been positive. Responses come back with a source URL and, where relevant, the underlying SEC document with a page number, which is exactly what we need for citations. The coverage is also aimed at the fields the open web doesn't expose cleanly, and the Fiscal.ai team was very responsive in getting us set up." - Miguel Aroca-Ouellette, Member of Technical Staff, Parallel Powering Search and agent workflows. Access to Fiscal.ai data is made possible through Parallel's Index product, where data and content owners can better understand how their properties are sourced in AI. Fiscal.ai is now live across two parts of the Parallel product: Parallel Search and Tasks. "It backs accurate fundamental financial queries in Parallel Search... It's also available as a tool to our agents on the Task API, so when a user asks a financial question (a company's adjusted EBITDA, the exact filing date of an annual report, etc.), the agent reaches for Fiscal instead of reading investor-relations pages. On that class of questions, it's well ahead of web search on both accuracy and citability." - Miguel Aroca-Ouellette, Member of Technical Staff, Parallel Supporting the next generation of financial AI. As more AI companies build agents that need to reason over financial markets, access to structured, citable data becomes increasingly important. Fiscal.ai is helping make that possible through modern, developer-friendly APIs. "There are now two users of the web with both Humans and Agents. Both require high quality financial data to support investment research and we are thrilled to partner with Parallel to combine two innovative products." - Braden Dennis, CEO, Fiscal.ai "We were impressed by Fiscal AI's flexible solution for distributing vital, high-quality financial data. Fiscal's model of processing and structuring fundamental data is both innovative and efficient, and we were excited to offer Fiscal's products to our customers." - Miguel Aroca-Ouellette, Member of Technical Staff, Parallel About Fiscal.ai. Fiscal.ai is the modern financial data company, building the infrastructure that powers the world's leading investors and platforms. Fiscal.AI deliver institutional-grade financial data with click-through auditability, available within minutes after filings, through Terminal, its all-in-one research platform; its Data Feed API, which powers institutions and investing platforms; and its MCP Connector, which brings accurate, source-linked data directly into AI tools. Its mission is to provide 100 million investors with high-quality financial data. About Parallel web systems. Parallel develops web infrastructure for AI agents. Its tools offer agents fresh, deep, and efficient context from the web from billions of pages and data providers, delivering the high-quality information agents need to perform real-world knowledge work. Get started with Parallel. Contact Fiscal.AI. A member of its team will reach out shortly.

Introducing the Search Capability Leaderboard. Today, Parallel is launching the Search Capability Leaderboard to help developers choose a language model for applications that use web search. At launch, Parallel is evaluating 24 popular open- and closed-source models from leading labs, comparing answer accuracy, cost, and cost efficiency. At Parallel, Parallel is constantly evaluating how model intelligence interacts with search to build and tune its APIs. While its proprietary and public evals suites are substantially more comprehensive than what Parallel is sharing today, Parallel hope this release encourages developers to further explore the interaction between model intelligence and web search. For more on that, read How to evaluate web search for AI. Comparing models with and without search. Using search well requires models to coordinate several distinct tasks: formulate effective queries, recognize when more evidence is needed, drill down into the right URLs, and synthesize the results into a correct answer. By holding search constant, Parallel can compare how different models handle those decisions and the tradeoffs developers face when choosing between them. In its experience, model labs don't consistently include dedicated web-search evaluations in model reports, while search provider comparisons focus on showcasing the capabilities of different search products. Developers choosing a model for search-heavy applications need to understand how accuracy and cost vary across models. What Parallel measure. Parallel focus on two measures to help developers compare models paired with search. The Search Intelligence Score looks at model accuracy across three existing public benchmark suites. The Search Efficiency ranking orders models by cost among those that meet or exceed the median accuracy score. Developers can use the accompanying data to compare models against their own accuracy requirements and budget. What its initial results show. Today's snapshot surfaces a few insights that developers may find interesting and actionable. Most results track the general capability rankings reported elsewhere, but a few show models gaining outsized benefit from search. Some takeaways: * DeepSeek V4.1 Flash gains 45 points with search, the most of any model; it ranks third for intelligence (65.3) and second for efficiency at $44.8 per 1,000 tasks. * GPT-6 (Astra) has the highest Search Intelligence Score, 70.3, at $301 per 1,000 tasks. * GPT-5.6 Luna leads on Search Efficiency, scoring 55.0 at $23.0 per 1,000 tasks, or 13x lower than Astra's cost. Whether higher accuracy justifies the additional cost depends on the application. These results make the tradeoff visible, helping developers decide which models deserve a closer look. Where Parallel go from here. Parallel hope the leaderboard is a helpful tool that will help developers in selecting models for their applications. Parallel plan to extend the coverage and methodology based on what developers find useful, and welcome feedback on the models, tasks, and comparisons that would help you make better decisions. By Parallel September 15, 2026