# Parallel vs. Perplexity

Openbenchmarks is an independent hub that publishes its harness and every raw vendor response. Its company search and coding-agent boards run both APIs with the same agent. On multi-turn company search, a fixed agent hunts for every company that matches three or four constraints. From search results alone, Parallel Basic scored 46.5 F1 against 37.8 for Perplexity. Its precision was 88.7 against 79.3, and its recall was 34.4 against 26.8.

Perplexity won on coding-agent tickets and on company search with page fetching. With search and fetch, it completed 77.7% of coding-agent documentation tickets against 77.0% for Parallel Advanced. When the company search agent could also fetch pages, Perplexity scored 46.6 F1 against 42.3 for Parallel Basic. Perplexity’s median search-only company task also cost less, $0.334 against $1.12. Openbenchmarks has flagged a timing bias on its company news board and is re-running every vendor, so this page leaves those results out until the re-run is published.

Parallel’s own September 2026 search runs gave both providers the same agent. With a GPT-5.6 Sol agent, Parallel Advanced scored 57.6 on WideSearch against 53.5 and 97% on SimpleQA Verified against 95%. Both scored 74% on BrowseComp, and Perplexity spent less to get there, $275 per 1,000 questions against $399.

OpenRouter’s August and September 2026 search benchmarks also favor Perplexity. Each run used a 25-turn search budget. With Claude Opus 5, Perplexity scored 89.0% on BrowseComp against 86.3% for Parallel, and 77.4% on Humanity’s Last Exam against 68.6%. With a GPT-5.6 Sol agent, it scored 84.0% on WideSearch against 80.6%. On DeepSearchQA the two were inside the margin of error, 77.0% for Parallel against 76.5%.

Parallel ran its Task API and Perplexity’s Agent API on the same fixed 100-question subsets on August 26, 2026. On BrowseComp, Task Lite scored 88% at $5 per 1,000 questions, and Task Core scored 91% at $25. Perplexity’s high effort level, as configured that day, scored 86% at $441.50. On DeepSearchQA, Task Lite scored 76% at $5 and Task Pro 83% at $100. Perplexity high scored 68% at $371.90.

Perplexity publishes higher numbers from its own harness. It reports BrowseComp scores of 78.0 for high and 85.33 for xhigh. It reports DeepSearchQA F1 of 89.6 for high and 93.0 for xhigh. F1 gives partial credit, so it is a different measure from the accuracy in Parallel’s run.

Every Task processor has a fixed price, from $5 to $2,400 per 1,000 runs. You know the cost of a run before it starts. Perplexity’s Agent API bills model tokens plus each tool call. The cost of a run depends on how many searches and fetches the agent makes.

Both APIs return structured output against a JSON Schema. Parallel also returns a Research Basis for each field. It holds the source URLs, the supporting excerpts, the reasoning, and a confidence rating. A pipeline can send low-confidence values to review and accept the rest. Perplexity’s Agent API cites sources inline in the answer, with no per-field confidence.

Task Groups run thousands of rows over one Task Spec and stream progress as runs complete. A finished run can post to your webhook. Perplexity’s background mode runs long jobs asynchronously, and you poll or stream each response by its id. Perplexity lists completion webhooks on its roadmap.

Parallel accepts 2,000 Task runs per minute on every account by default. Perplexity’s Agent API allows 1 to 33 requests per second, set by lifetime spend. That is 60 per minute for a new account and 1,980 per minute after $1,000 in purchases.

Monitor runs a query on a schedule and posts new events to your webhook. You pick hourly, daily, or weekly checks. You define the query once and skip the cron jobs and deduplication. Monitor costs $3 to $10 per 1,000 checks. Perplexity documents no monitoring endpoint. Its cookbook builds a news monitor from repeated Search API calls.

FindAll turns a description into a verified, structured list of matching entities. Each run costs a fixed fee plus $0.03 to $1 per match, set by the generator you pick. Entity Search returns candidate entities in one call. Perplexity offers people search and finance search as Agent API tools. It has no endpoint that builds a verified entity list.

Parallel serves its Search and Extract APIs from the EU at eu.parallel.ai, under an enterprise agreement. The API, SDKs, and keys are the same as api.parallel.ai, and only the base URL changes. Search runs in basic mode only, with up to 20 results per request, answered from the EU index with no live fetching. Extract takes up to 20 URLs per request. Queries, objectives, results, excerpts, and model inference stay in the EU, and request and response content is not retained.

The EU endpoint has limits. Billing, metrics, and edge logs, which contain no content, go to the US. Extract looks up each URL in Parallel’s cache of public web pages, which is hosted outside the EU. Only the URL is sent, and under zero data retention it isn’t saved. Extract is cache-only when advanced_settings.fetch_policy.timeout_seconds is 0, and uncached URLs then return an error. Other Search modes, Task, FindAll, Monitor, Chat, Responses, and MCP aren’t served in the EU. They return an error and are never forwarded to the US. Our guide to EU search APIs compares every vendor’s documented option.

Perplexity doesn’t document an EU processing option. Its API FAQ says compute is hosted on Amazon Web Services in North America. By default the API keeps no prompt data, so retention is covered but location isn’t. Perplexity’s position was read from its own documentation on October 16, 2026.