Search API # The best web search for your AI agent , model, IDE, chatbot, application, workflow The highest accuracy web search API, built from the ground up for AIs [Get Started ](https://platform.parallel.ai/)[Get a Demo](https://contact.parallel.ai/) Why Parallel ## An agent is only as good as its context Parallel returns the best information from the web ## Declare semantic objectives, not just keywords AI tells Parallel Search exactly what it's looking for ## Get back URLs ranked for token relevancy Parallel surfaces the most information-dense pages for the agent's next action ## Reason on compressed token efficient excerpts Each URL is distilled into the highest-value tokens for optimal context windows ## We optimize every web search token in the context window This means agent responses are more accurate and cost less [Search Playground](https://platform.parallel.ai/) ## HLE Search | Series | Model | Cost (CPM) | Accuracy (%) | | --------- | ------------ | ----------- | ------------ | | Parallel | parallel | 82 | 47 | | Others | exa | 138 | 24 | | Others | tavily | 190 | 21 | | Others | perplexity | 126 | 30 | | Others | openai gpt-5 | 143 | 45 | ### About this benchmark This [benchmark](https://lastexam.ai/) consists of 2,500 questions developed by subject-matter experts across dozens of subjects (e.g. math, humanities, natural sciences). Each question has a known solution that is unambiguous and easily verifiable, but requires sophisticated web retrieval and reasoning. Results are reported on a sample of 100 questions from this benchmark. ### Methodology * **Evaluation**: Results are based on tests run using official Search MCP servers provided as an MCP tool to OpenAI's GPT-5 model using the Responses API. In all cases, the MCP tools were limited to only the appropriate web search tool. Answers were evaluated using an LLM as a judge (GPT 4.1). * **Cost Calculation**: Cost reflects the average cost per query across all questions run. This cost includes both the search API call and LLM token cost. * **Testing Dates**: Testing was conducted from November 3rd to November 5th, 2025. ## BrowseComp Search | Series | Model | Cost (CPM) | Accuracy (%) | | --------- | ------------ | ----------- | ------------ | | Parallel | parallel | 156 | 58 | | Others | exa | 233 | 29 | | Others | tavily | 314 | 23 | | Others | perplexity | 256 | 22 | | Others | openai gpt-5 | 253 | 53 | ### About this benchmark This [benchmark](https://openai.com/index/browsecomp/), created by OpenAI, contains 1,266 questions requiring multi-hop reasoning, creative search formulation, and synthesis of contextual clues across time periods. Results are reported on a sample of 100 questions from this benchmark. ### Methodology * **Evaluation**: Results are based on tests run using official Search MCP servers provided as an MCP tool to OpenAI's GPT-5 model using the Responses API. In all cases, the MCP tools were limited to only the appropriate web search tool. Answers were evaluated using an LLM as a judge (GPT 4.1). * **Cost Calculation**: Cost reflects the average cost per query across all questions run. This cost includes both the search API call and LLM token cost. * **Testing Dates**: Testing was conducted from November 3rd to November 5th, 2025. ## WebWalker-Search | Series | Model | Cost (CPM) | Accuracy (%) | | --------- | ------------ | ----------- | ------------ | | Parallel | parallel | 42 | 81 | | Others | exa | 107 | 48 | | Others | tavily | 156 | 79 | | Others | perplexity | 91 | 67 | | Others | openai gpt-5 | 88 | 73 | ### About this benchmark This [benchmark](https://arxiv.org/abs/2501.07572) is designed to assess the ability of LLMs to perform web traversal. To successfully answer the questions in the benchmark, it requires the ability to crawl and extract content from website subpages. Results are reported on a sample of 100 questions from this benchmark. ### Methodology * **Evaluation**: Results are based on tests run using official Search MCP servers provided as an MCP tool to OpenAI's GPT-5 model using the Responses API. In all cases, the MCP tools were limited to only the appropriate web search tool. Answers were evaluated using an LLM as a judge (GPT 4.1). * **Cost Calculation**: Cost reflects the average cost per query across all questions run. This cost includes both the search API call and LLM token cost. * **Testing Dates**: Testing was conducted from November 3rd to November 5th, 2025. ## FRAMES-Search | Series | Model | Cost (CPM) | Accuracy (%) | | --------- | ------------ | ----------- | ------------ | | Parallel | parallel | 42 | 92 | | Others | exa | 81 | 81 | | Others | tavily | 122 | 87 | | Others | perplexity | 95 | 83 | | Others | openai gpt-5 | 68 | 90 | ### About this benchmark This [benchmark](https://huggingface.co/datasets/google/frames-benchmark) contains 824 challenging multi-hop questions designed to test factuality, retrieval accuracy, and reasoning. Results are reported on a sample of 100 questions from this benchmark. ### Methodology * **Evaluation**: Results are based on tests run using official Search MCP servers provided as an MCP tool to OpenAI's GPT-5 model using the Responses API. In all cases, the MCP tools were limited to only the appropriate web search tool. Answers were evaluated using an LLM as a judge (GPT 4.1). * **Cost Calculation**: Cost reflects the average cost per query across all questions run. This cost includes both the search API call and LLM token cost. * **Testing Dates**: Testing was conducted from November 3rd to November 5th, 2025. ## Batched SimpleQA - Search | Series | Model | Cost (CPM) | Accuracy (%) | | --------- | ------------ | ----------- | ------------ | | Parallel | parallel | 50 | 90 | | Others | exa | 119 | 71 | | Others | tavily | 227 | 59 | | Others | perplexity | 100 | 74 | | Others | openai gpt-5 | 91 | 88 | ### About this benchmark This benchmark was created by batching 3 independent questions from the original [SimpleQA dataset](https://openai.com/index/introducing-simpleqa/) to create 100 composite, more complex, questions. ### Methodology * **Evaluation**: Results are based on tests run using official Search MCP servers provided as an MCP tool to OpenAI's GPT-5 model using the Responses API. In all cases, the MCP tools were limited to only the appropriate web search tool. Answers were evaluated using an LLM as a judge (GPT 4.1). * **Cost Calculation**: Cost reflects the average cost per query across all questions run. This cost includes both the search API call and LLM token cost. * **Testing Dates**: Testing was conducted from November 3rd to November 5th, 2025. ## SimpleQA Search | Series | Model | Cost (CPM) | Accuracy (%) | | --------- | ------------ | ----------- | ------------ | | Parallel | parallel | 17 | 98 | | Others | exa | 57 | 87 | | Others | tavily | 110 | 93 | | Others | perplexity | 52 | 92 | | Others | openai gpt-5 | 37 | 98 | ### About this benchmark This [benchmark](https://openai.com/index/introducing-simpleqa/), created by OpenAI, contains 4,326 questions focused on short, fact-seeking queries across a variety of domains. Results are reported on a sample of 100 questions from this benchmark. ### Methodology * **Evaluation**: Results are based on tests run using official Search MCP servers provided as an MCP tool to OpenAI's GPT-5 model using the Responses API. In all cases, the MCP tools were limited to only the appropriate web search tool. Answers were evaluated using an LLM as a judge (GPT 4.1). * **Cost Calculation**: Cost reflects the average cost per query across all questions run. This cost includes both the search API call and LLM token cost. * **Testing Dates**: Testing was conducted from November 3rd to November 5th, 2025. # Powered by our own proprietary web scale index With innovations in retrieval, crawling, indexing, and reasoning * Billions of pages covering the full depth and breadth of the public web * Millions of pages added daily * Intelligently recrawled to keep data fresh # The knowledge of the entire public web in a single tool call Integrated directly, or add our [MCP Server](https://docs.parallel.ai/integrations/mcp/programmatic-use) [Search Playground](https://platform.parallel.ai/)[Docs](https://docs.parallel.ai/search/search-quickstart) ## Scale with unmatched price-performance Get started with up to 80,000 free search requests $0.001 per request with 10 results + $0.001 per page extracted [Get Started ](https://platform.parallel.ai/)[Calculate Savings](https://parallel.ai/products/search/calculator) | | Search API | | ----------------- | ----------------------------------- | | Inputs | Search objective, Keywords | | Outputs | Ranked URLs, Compressed excerpts | | Best for | Web search tool calls for AI agents | | Latency | 200ms - 3s, synchronous | | Basis | — | | Rate limits | 600 requests / min | | Security | SOC2 | | Price per request | $0.001 - $0.005 for 10 results | ## Every control you need across any web page Premium content extraction Fetch content from PDFs and sites that are JS heavy or have CAPTCHAs Freshness policies Set page age triggers for live crawls, with timeout thresholds to gaurantee latency LLM friendly outputs Choose between dense snippets or full page contents, in markdown LLMs understand Source control Pick which domains are included or excluded from your web search results # Secure and trusted Zero data retention Soc 2 Type 2 No training ## FAQ +−What is the Parallel Search API? Parallel Search (API) is the highest accuracy AI search API. It allows developers to build AI apps, agents, and workflows that can search for and retrieve data from the web. It can be integrated into agent workflows for deep research across multiple steps, or for more basic single-hop queries. +−What is declarative semantic search? Declarative semantic search lets agents express intent in natural language rather than construct keyword queries. Instead of "Columbus" AND "corporate law" AND "disability", an agent specifies: "Columbus-based corporate law firms specializing in disability care." The Search API interprets meaning and context, not just keywords, making it natural to integrate into agent workflows where you already have rich context from previous reasoning steps. +−What makes Parallel different from other search providers? Parallel is the only Search API built from the ground up for AI agents. This means that agents can specify declarative semantic objectives and Parallel returns URLs and compressed excerpts based on token relevancy. The result is extremely dense web tokens optimized to engineer your agent’s context for better reasoning at the next turn. Agents using Parallel search produce answers with higher accuracy, fewer round trips, and lower cost. +−Where do search results come from? How fresh are they? We maintain a large web index containing billions of pages. Our crawling, retrieval, and ranking systems add and update millions of pages daily to keep the index fresh. +−Does Parallel have a web crawler? Yes, Parallel operates a web crawler to support the quality and coverage of the index. Our crawler respects _\_robots.txt\__ and related crawling directives. [Learn more about Parallel’s crawler here](https://docs.parallel.ai/resources/crawler). +−What are dense excerpts? Dense excerpts are the most query relevant content from a webpage, compressed to be extremely token efficient for an agent. These compressed excerpts reduce noise by engineering an agent’s context window to only have the most relevant tokens to reason on - leading to higher accuracy, fewer round trips, and less token use. +−What does end-to-end latency mean? End-to-end latency measures total time from agent input to final output, not single-search latency. Our semantic search architecture and dense snippets reduce the number of searches required to reach quality outputs. Two high-precision searches with Parallel beat three lower-quality attempts elsewhere—saving both time and tokens.