Oct 27, 2026
- [How Bretton runs AI compliance, fraud, and risk investigations with 3x faster web research using Parallel](https://parallel.ai/blog/case-study-bretton-ai)
Tags:Customers
Author: By ParallelOctober 29, 2026
# Parallel: Unmatched SOTA web search for Finance AI Agents
Parallel for Finance is a universal upgrade to our index and ranking for financial agents. On FinSearchComp's hardest tier, Parallel is the most accurate search API, at up to 2x lower cost than Exa and Perplexity, and on our private finance benchmark, it matches OpenAI's native search for 30% to 40% less.
Tags:Product
Finance agents need today's filings, registers, and news at inference time, and the facts that matter most often sit outside licensed datasets: in regulator registers, local-exchange filings, and private-company announcements.
Today we're launching Parallel for Finance, a definitive upgrade to our index and ranking for financial agents across all of Parallel's APIs, with no new plan or subscription. Key takeaways:
- - On FinSearchComp[FinSearchComp]'s hardest tier (T3), Parallel is the most accurate search API at up to **2x lower cost than Exa and Perplexity**. Parallel Fast comes second, at 10x lower cost than the** others.**
- - On Samaya's FrontierFinance[FrontierFinance], Parallel ties for the top score at **38% lower cost than Exa**, and **beats OpenAI, Perplexity, and Tavily on both accuracy and cost.**
- - On our private PWS-Finance benchmark, Parallel leads, **beating OpenAI's native search at 30% lower cost**. Rogo[Rogo], Kepler[Kepler], Bretton[Bretton], and Taktile already run Parallel in production.
## Defining the cost-accuracy frontier for finance
We ran every search API in the same agent loop with the same questions and the same judge, so the search layer is the only thing that changes. Cost is end-to-end at list price: the answer model's tokens plus every search and fetch the agent makes. The charts cover FinSearchComp, Samaya's FrontierFinance[FrontierFinance] for long-form investment research, and PWS-Finance, our private benchmark. Methodology for each is in the notes under its chart.
FinSearchComp T3FinSearchComp T2FrontierFinancePWS-Finance
Correct answers per $
Parallel Fast35.3/$ | 53.7% / 15.2CPM
Parallel Advanced3.6/$ | 57.4% / 161.1CPM
OpenAI native search2.4/$ | 48.3% / 204.3CPM
Exa Auto2.3/$ | 51.4% / 226.2CPM
Perplexity Finance + Web1.4/$ | 50% / 347CPM
Tavily0.9/$ | 43% / 481.8CPM
COST (CPM)
ACCURACY (%)
Loading chart...
CPM: USD per 1,000 questions, end to end (answer-model tokens plus search and fetch calls at list price).
Parallel
Others
Benchmark comparison across Cost (CPM) and Accuracy (%). CPM: USD per 1,000 questions, end to end (answer-model tokens plus search and fetch calls at list price).
+Methodology
**Dataset**
FinSearchComp[FinSearchComp] has 635 questions, written by 70 financial experts, that reproduce analyst work across global and Greater China markets. T3 is the complex historical investigation tier, the hardest in the benchmark; results are reported on the 172 graded questions.
**Evaluation methodology**
Multi-step agentic evaluation: every search API runs in the same agent loop with the same questions and the same LLM judge, so the search layer is the only variable. The agent calls the provider's search tool plus its fetch tool where one exists. Parallel Advanced, Exa Auto, OpenAI native search, Perplexity and Tavily answer with GPT-6 Sol (reasoning: high); Parallel Fast runs as the low-latency stack with GPT-6 Luna (reasoning: high). Scores are the mean of three runs unless noted. The benchmark's answers are published online. For the other search APIs, our harness drops those copies before the model sees them. OpenAI native search answers that came from published copies count as wrong (OpenAI native search scores 64.0 before that adjustment). Tavily is a single run.
Cost is end-to-end per answer at list price: the answer model's tokens plus every search and fetch the agent makes, shown per 1,000 questions.
**Testing dates**
Evals were run on October 22, 2026 (Tavily on October 28, 2026).
**Dataset**
FinSearchComp[FinSearchComp] has 635 questions, written by 70 financial experts, that reproduce analyst work across global and Greater China markets. T3 is the complex historical investigation tier, the hardest in the benchmark; results are reported on the 172 graded questions.
**Evaluation methodology**
Multi-step agentic evaluation: every search API runs in the same agent loop with the same questions and the same LLM judge, so the search layer is the only variable. The agent calls the provider's search tool plus its fetch tool where one exists. Parallel Advanced, Exa Auto, OpenAI native search, Perplexity and Tavily answer with GPT-6 Sol (reasoning: high); Parallel Fast runs as the low-latency stack with GPT-6 Luna (reasoning: high). Scores are the mean of three runs unless noted. The benchmark's answers are published online. For the other search APIs, our harness drops those copies before the model sees them. OpenAI native search answers that came from published copies count as wrong (OpenAI native search scores 64.0 before that adjustment). Tavily is a single run.
Cost is end-to-end per answer at list price: the answer model's tokens plus every search and fetch the agent makes, shown per 1,000 questions.
**Testing dates**
Evals were run on October 22, 2026 (Tavily on October 28, 2026).
### FinSearchComp T3 (Parallel for Finance, Oct 2026)
| Series | Model | Cost (CPM) | Accuracy (%) | | -------- | ------------------------ | ---------- | ------------ | | Parallel | Parallel Fast | 15.2 | 53.7 | | Parallel | Parallel Advanced | 161.1 | 57.4 | | Others | Exa Auto | 226.2 | 51.4 | | Others | OpenAI native search | 204.3 | 48.3 | | Others | Perplexity Finance + Web | 347 | 50 | | Others | Tavily | 481.8 | 43 |
CPM: USD per 1,000 questions, end to end (answer-model tokens plus search and fetch calls at list price).
**Dataset**
FinSearchComp[FinSearchComp] has 635 questions, written by 70 financial experts, that reproduce analyst work across global and Greater China markets. T3 is the complex historical investigation tier, the hardest in the benchmark; results are reported on the 172 graded questions.
**Evaluation methodology**
Multi-step agentic evaluation: every search API runs in the same agent loop with the same questions and the same LLM judge, so the search layer is the only variable. The agent calls the provider's search tool plus its fetch tool where one exists. Parallel Advanced, Exa Auto, OpenAI native search, Perplexity and Tavily answer with GPT-6 Sol (reasoning: high); Parallel Fast runs as the low-latency stack with GPT-6 Luna (reasoning: high). Scores are the mean of three runs unless noted. The benchmark's answers are published online. For the other search APIs, our harness drops those copies before the model sees them. OpenAI native search answers that came from published copies count as wrong (OpenAI native search scores 64.0 before that adjustment). Tavily is a single run.
Cost is end-to-end per answer at list price: the answer model's tokens plus every search and fetch the agent makes, shown per 1,000 questions.
**Testing dates**
Evals were run on October 22, 2026 (Tavily on October 28, 2026).
### FinSearchComp T2 (Parallel for Finance, Oct 2026)
| Series | Model | Cost (CPM) | Accuracy (%) | | -------- | ------------------------ | ---------- | ------------ | | Parallel | Parallel Fast | 3.3 | 77.5 | | Parallel | Parallel Advanced | 35.1 | 79.9 | | Others | Exa Auto | 69.9 | 77.9 | | Others | OpenAI native search | 70.2 | 77.9 | | Others | Perplexity Finance + Web | 104.6 | 79.7 | | Others | Tavily | 169 | 77.2 |
CPM: USD per 1,000 questions, end to end (answer-model tokens plus search and fetch calls at list price).
**Dataset**
FinSearchComp[FinSearchComp] has 635 questions, written by 70 financial experts, that reproduce analyst work across global and Greater China markets. T2 is the simple historical lookup tier; results are reported on the 219 graded questions.
**Evaluation methodology**
Multi-step agentic evaluation: every search API runs in the same agent loop with the same questions and the same LLM judge, so the search layer is the only variable. The agent calls the provider's search tool plus its fetch tool where one exists. Parallel Advanced, Exa Auto, OpenAI native search, Perplexity and Tavily answer with GPT-6 Sol (reasoning: high); Parallel Fast runs as the low-latency stack with GPT-6 Luna (reasoning: high). Scores are the mean of three runs unless noted. The benchmark's answers are published online. For the other search APIs, our harness drops those copies before the model sees them. OpenAI native search answers that came from published copies count as wrong (OpenAI native search scores 82.8 before that adjustment). Tavily is a single run.
Cost is end-to-end per answer at list price: the answer model's tokens plus every search and fetch the agent makes, shown per 1,000 questions.
**Testing dates**
Evals were run on October 22, 2026 (Tavily on October 28, 2026).
### FrontierFinance (Parallel for Finance, Oct 2026)
| Series | Model | Cost (CPM) | Score (%) | | -------- | --------------------- | ---------- | --------- | | Parallel | Parallel Fast | 201.5 | 46 | | Parallel | Parallel Advanced | 234.2 | 46.8 | | Others | Exa Auto | 376.3 | 47.2 | | Others | OpenAI native search | 298.7 | 45 | | Others | Perplexity Search API | 254.3 | 44.3 | | Others | Tavily | 470.3 | 42.1 |
CPM: USD per 1,000 questions, end to end (answer-model tokens plus search and fetch calls at list price).
**Dataset**
Samaya's FrontierFinance[FrontierFinance] has 220 expert-written research queries covering the investment workflow from screening and company research to earnings, modeling and catalyst monitoring. Answers are long-form; a panel of three judges scores each one against expert rubrics, and the score is the rubric pass rate by judge majority.
**Evaluation methodology**
Every search API runs in the same agent loop with the same questions and the same judges, with a budget of 20 tool calls per question. All arms, including Parallel Fast, answer with GPT-6 Sol (reasoning: high). Parallel runs through its hosted search and fetch tools; Exa, OpenAI and Tavily use search plus fetch; Perplexity is the Search API only, with no fetch tool. One run per arm.
Cost is end-to-end per answer at list price: the answer model's tokens plus every search and fetch the agent makes, shown per 1,000 questions.
**Testing dates**
Evals were run on October 27, 2026.
### PWS-Finance (Parallel for Finance, Oct 2026)
| Series | Model | Cost (CPM) | Accuracy (%) | | -------- | --------------------- | ---------- | ------------ | | Parallel | Parallel Fast | 110.3 | 83 | | Parallel | Parallel Advanced | 125.5 | 84.3 | | Others | Exa Auto | 197.6 | 75.9 | | Others | OpenAI native search | 180.7 | 83.7 | | Others | Perplexity Search API | 121.9 | 74.3 |
CPM: USD per 1,000 questions, end to end (answer-model tokens plus search and fetch calls at list price).
**Dataset**
PWS-Finance is Parallel's private finance benchmark: 174 held-out questions across eight finance workflows (company profiles, funding, deals, ownership, public financials, regulation, markets and macro, developments), each asking for specific facts about one entity as of one date. The questions and answers are not published anywhere online.
**Evaluation methodology**
Every search API runs in the same agent loop with the same questions and the same LLM judge, with a budget of 20 tool calls per question. All arms, including Parallel Fast, answer with GPT-6 Sol (reasoning: high). Strict scoring: an answer is correct only if every required fact is right. Perplexity is the Search API only, not Finance + Web. Tavily is not plotted: it ran with the model's date filters on (98% of searches limited to the past year) and hit the 20-call cap on 77% of questions, so its 36.4% at $432.10 CPM is not comparable to the other arms. Scores are the mean of three runs.
Cost is end-to-end per answer at list price: the answer model's tokens plus every search and fetch the agent makes, shown per 1,000 questions.
**Testing dates**
Evals were run on October 27, 2026.
### FinSearchComp:
FinSearchComp[FinSearchComp] has 635 questions written by 70 financial experts to reproduce analyst work across global and Greater China markets. We ran its two historical tiers: simple lookups (T2) and complex multi-step investigations (T3).
On T3, Parallel Advanced is the most accurate search API and cheaper than every competitor, at 1.3x to 3x lower cost. Parallel Fast beats every competitor at 13x+ lower cost. On T2, Parallel Advanced posts the top score at half the cost of the cheapest alternative, and Parallel Fast comes within half a point of it at 20x lower cost.
### PWS-Finance private benchmark: Beats OpenAI at 30% lower cost
Public benchmarks and their answers end up online and in training data. To account for "benchmaxxing" bias, we gathered 174 private questions across eight finance workflows informed by real financial AI uses currently in production.
Parallel Advanced posts the top score, beating OpenAI's native search at 30% lower cost and Exa and Perplexity by 8 to 10 points. Parallel Fast comes within a point of OpenAI at 39% lower cost.
## How we built the benchmark
Vendor benchmarks tend to favor the vendor that builds them. To avoid that, we sourced every question from real research demand rather than Parallel's strengths, gold-labeled answers using each competitor's highest agentic mode, and ran every provider through the same harness with its published best practices, so only the search provider changes.
Each question asks for specific facts about one entity as of one date.
- - **Detailed questions.** The eight workflows are company profiles, funding and investors, deals, ownership and leadership, public financials, regulation and compliance, markets and macro, and recent developments.
- - **Global coverage.** The question set is global. There are 65 questions about the US, 34 about Europe, 24 about Greater China, and the rest cover Asia-Pacific, the Middle East, Africa and Latin America.
- - **Challenging questions.** Most questions carry a trap that a quick search gets wrong. The facts may be spread across several sources, published months after the period they cover, or different depending on which version of a filing you read.
- - **Verified answers.** Each answer key draws on several independent research agents, competitors' as well as our own. Critics and a verifier then check the results, and these checks catch 93% to 100% of deliberately planted errors.
Here are four sample questions and answers from the set:
| Workflow | Question | Answer |
|---|---|---|
| Regulation and compliance | As of 22 October 2026, which banking legal entities in the Abu Dhabi Commercial Bank PJSC group held an active authorisation shown on the UK Financial Conduct Authority's Financial Services Register? Give the count and, for each entity, its legal name, firm reference number, authorisation or permission type, and status. | None. No entity in the ADCB group holds an active FCA authorisation. The group's former UK representative office was dissolved in April 2024, and a representative office is not an authorised firm in any case. |
| Ownership and people | As of 21 October 2026, how many subsidiaries does ArcelorMittal's 2025 annual report list under Poland, and what are up to five of their names? | One: ArcelorMittal Poland S.A., held at 100%, in the list of principal operating subsidiaries in the 2025 Form 20-F. |
| Deals and transactions | As of October 23, 2026, which acquisitions of businesses did Rush Enterprises, Inc. complete during calendar year 2025, excluding stand-alone asset purchases? For each, give the target, closing date, and disclosed purchase price, or “not disclosed” if no price was disclosed. | One: Leeds Transit, Inc., a Canadian bus dealership group, closed June 16, 2025, for about US$25.6 million. The price appears in note 15 of the FY2025 Form 10-K; the June 2025 press release did not disclose it. |
| Markets and macro | As of 22 October 2026, what reference month and release date did Banque de France's latest monthly balance-of-payments release give, and what were France's seasonally and working-day adjusted current-account balance for that month and its revised balance for the preceding month, in EUR billions? | August 2026, released October 7, 2026. Current-account deficit of EUR 1.5 billion for August; July revised to a EUR 3.9 billion deficit, from the EUR 4.7 billion first published. |
## How teams deploy Parallel for Finance in production
Agents built on Parallel already produce tear sheets, run diligence and KYB checks, and monitor portfolios across private companies, funds, and non-US markets. They cover everything from a quick lookup to deep investigations. Here is how teams run Parallel in production today.
Rogo
1 of 3
An AI analyst for investment banks
Hundreds of thousands of requests a day
Rogo runs its agents on Parallel Search and Extract across filings, earnings calls and market sources. At peak, that's hundreds of thousands of requests a day.
“Of all the providers we tested, Parallel delivered the highest quality results at the best price.” Gabriel Stengel, CEO, Rogo
## Build your first tear sheet
Our open-source cookbook shows how to use Parallel's APIs to generate end-to-end tear sheets outlining investment cases for a dataset of companies. From a CSV of tickers, the recipe runs the Task API to research each company's profile, four years of financials, market data, recent news, and a bull and bear case, then renders a cited two-page tear sheet plus a research basis page listing every source. It's intended as a lightweight illustration of the workflows we have seen work well in production. The code is in our open-source repo[our open-source repo].
## Get started
Start building for free[Start building for free] or dive into our docs[docs].
Contact us[Contact us] to learn more about enterprise plans with ZDR, EU data residency, SSO, and more.
