--- title: "Parallel AI API: pricing, rate limits & MCP | ONEGTMLAB" description: "Parallel AI API: Pay-as-you-go web APIs for AI agents on Parallel's own index: Search, Extract, Task (deep research and structured enrichment with…" canonical: https://www.onegtmlab.dev/tool-directory/research-and-scraping/parallel-ai --- # Parallel AI API: pricing, rate limits & MCP | ONEGTMLAB [](/) [ Book a call ](/book) ![](/_astro/parallel-ai.DNKGhCFf.svg) API review # Parallel AI API review Pay-as-you-go web APIs for AI agents on Parallel's own index: Search, Extract, Task (deep research and structured enrichment with citations), Responses, Monitor, FindAll and Entity Search, plus a free anonymous Search MCP. Best for row-level account/contact research and enrichment with per-field citations and confidence. Verified 25 Oct 2026 · 46 endpoints listed · Official MCP server [ See pricing ](#pricing)[ Parallel AI API docs ](https://docs.parallel.ai/api-reference/search/search) At a glance Category Research, Scraping, Enrichment, Sourcing, Signals Pricing model Usage-based Free tier PUBLIC PAGES: 'Run up to 5,000 requests per month for free' and 'Earn up to $80… Auth API key Official MCP Yes Best for Row-level account/contact research and enrichment with per-field citations and confidence [parallel.ai ↗](https://parallel.ai)[Pricing page ↗](https://docs.parallel.ai/getting-started/pricing) ## What a call costs The actions you pay for, as Parallel AI prices them. Action Cost Per Search turbo $1 1k requests (10 results); +$1 per 1k additional results Search fast $1 1k requests (10 results); +$1 per 1k additional results Search basic $5 1k requests (10 results); +$1 per 1k additional results Search advanced (API default mode) $5 1k requests (10 results); +$1 per 1k additional results Image Search fast / advanced $1 / $5 1k requests (up to 20 results) Extract $1 1k URLs (docs) = '1,000 results' (pricing page and in-app rate card); measured 1 URL = $0.001 Task lite $5 1k successful runs Task base $10 1k successful runs Show all 25 priced actionsHide the rest Action Cost Per Task core $25 1k successful runs Task core2x $50 1k successful runs Task pro $100 1k successful runs Task ultra $300 1k successful runs Task ultra2x $600 1k successful runs Task ultra4x $1,200 1k successful runs Task ultra8x $2,400 1k successful runs Responses low / medium (default) / high $10 / $50 / $250 1k successful requests Chat API (beta, legacy page) $5 1k requests Monitor lite / base $3 / $10 1k executions (checks) FindAll preview $0.10 fixed + $0.00/match run FindAll base $0.25 fixed + $0.03/match run FindAll core $2.00 fixed + $0.15/match run FindAll pro $10.00 fixed + $1.00/match run FindAll enrichment Task processor price per match per enrichment Entity Search $5 per 1k requests (100 results) + $0.05 per 1k additional results request Data Connectors pay-per-use per successful invocation (provider-specific); bioRxiv, ClinicalTrials.gov, ChEMBL, CMS Coverage, medRxiv, NPI Registry, PubMed free invocation ## Pricing, plan by plan US dollars, as Parallel AI's pricing page shows them on 25 Oct 2026. Monthly and annual billing are shown apart. Free tier PUBLIC PAGES: 'Run up to 5,000 requests per month for free' and 'Earn up to $80 at signup + $5 in free credits per month'; 'up to 80,000 free search requests'; $5/month needs a card, expires month end, one org per card, marketplace/postpaid excluded. LIVE APP (new org, 2026-10-25): plan named 'Free', 'No card needed'; the full $80.00 signup credit landed at once with no card or survey and 'Expires Dec 24' (about 60 days); 'Add a card to unlock monthly credits'; 'Requests stop when your credit runs out'. The live app wins for what a new user gets: $80 immediately, expiring ~60 days after signup. Startups: up to $250 by application. Search MCP free without an account. ### Pay as you go One-off pack Prepaid dollar balance, priced per request; optional auto-reload API includedPlan details Billed at standard rates beyond free credit; monthly spend limits are notify-only and never block requests ### Free $0 Free $80 signup credit granted immediately, no card, expires ~60 days after signup (in-app 'Expires Dec 24' for a 25 Oct signup); +$5/month once a card is added API includedPlan details Requests stop when credit runs out unless a card is added; then standard rates ### Startup program [Price not verified](https://docs.parallel.ai/getting-started/pricing) Up to $250 in free credits for qualified startups (application) API included ### Enterprise Custom quote Custom; ZDR, DPA, SSO, custom rate limits, EU data residency API included ## Rate limits Defaults per product, rolling 60-second window, only POSTs that create resources count (GETs and polling are free of limits). Limit Search (POST /v1/search) Image Search Extract Tasks / Task Groups (per run created) Chat (/v1beta/chat/completions) Responses FindAll run creation Entity Search Monitor creation Search MCP anonymous Enterprise Limit 600 per min 600 per min 600 per min 2,000 per min 300 per min 300 requests / min (pricing page only; not on rate-limits page) 300 per hour (docs) vs 25 / hr (pricing page): conflicting 600 per min 300 per min 'lower rate limits', exact cap not published; API key or OAuth unlocks higher Custom rate limits More on limitsLess 429 on breach. Higher limits No rate-limit headers are returned (only x-request-id), per hands-on. ## How to get an API key Parallel AI uses API key. The steps below follow Parallel AI's own docs. 1. Create a Parallel AI account. The Free plan includes API access. 2. Sign up at [platform.parallel.ai](https://platform.parallel.ai) (magic-link email or Google; SSO on login; no password, no card, no captcha). A Default App + Default API Key are created at signup; more under Settings > API Keys, grouped by App. Full keys stay re-viewable by admins (eye icon), and per-user 'Account Keys' are created automatically for Platform/CLI sign-ins. 3. Send it with every request: header x-api-key: for every product endpoint (spec ApiKeyAuth; the EU host eu.parallel.ai takes the same header). Account API (`https://api.parallel.ai/account/`...) needs Authorization: Bearer minted by Parallel OAuth (device / jwt-bearer grants); an API key gets 401 'Missing bearer token' (live test). MCP servers take Authorization: Bearer . 4. Make a first call: Web search returning ranked URLs with LLM-ready compressed excerpts (GA). Keep the key in an environment variable, never in the command. ``` curl -X POST "https://api.parallel.ai/v1/search" \ -H "Authorization: Bearer $PARALLEL_AI_API_KEY" \ -H "Content-Type: application/json" \ -d '{}' ``` ## Where the docs live What Where Documentation [docs.parallel.ai](https://docs.parallel.ai) API reference [docs.parallel.ai/api-reference/search/search](https://docs.parallel.ai/api-reference/search/search) OpenAPI spec [docs.parallel.ai/public-openapi.json](https://docs.parallel.ai/public-openapi.json) llms.txt [docs.parallel.ai/llms.txt](https://docs.parallel.ai/llms.txt) Changelog [docs.parallel.ai/resources/changelog](https://docs.parallel.ai/resources/changelog) Base URL `https://api.parallel.ai` MCP server `https://search.parallel.ai/mcp` SDKs parallel-web (Python, PyPI), parallel-web (TypeScript, npm), parallel-cli (pipx/uv 'parallel-web-tools\[cli\]' or brew parallel-web/tap/parallel-cli), parallel-web-tools (Python data integrations: Polars, DuckDB, Spark), OpenAI SDKs (Responses and Chat are OpenAI-compatible), Parallel Agent Skills (npx skills add parallel-web/parallel-agent-skills), Claude Code plugin, Cursor plugin, OpenCode plugin, Pi extension ## Strengths and limits ### Where it fits - Row-level account/contact research and enrichment with per-field citations and confidence (Task API, $5-$2,400 per 1k runs) - Batch enrichment of lists via Task Groups (up to 1,000 runs per request) - Building verified entity lists from a natural-language ICP (FindAll, $0.25 + $0.03/match on base) - Cheap agent web search ($1 per 1k on turbo/fast) and URL-to-markdown extraction ($1 per 1k URLs) - Scheduled web monitoring for signals with webhooks (Monitor, $3 per 1k executions on lite) - Free zero-setup web search inside Claude Code / Cursor via the anonymous Search MCP ### Where it falls short - Pages behind a login (public web only, per FAQ) - Verified emails/phones or static firmographic databases - Reselling outputs or building a database / data-selling product from outputs (Customer Terms 2(c)(vi)) - Production-critical FindAll / Entity Search pipelines that cannot absorb breaking changes (public beta, 30 days notice) ### Worth knowing - Search API default mode is 'advanced' at $5 per 1k; 'fast' is $1 per 1k and the docs call it the right choice for most agents. Set mode explicitly. - Search prices include 10 results; every extra result costs $1 per 1k, so max\_results=50 on basic/advanced is $0.045 per call, not $0.005. - Task is billed per successful run regardless of field count, but processors cap fields (~2 lite, ~5 base, ~10 core/core2x, ~20 pro/ultra, ~25 ultra2x/4x/8x per pricing page and in-app); spec+input over 25,000 chars errors. - Prepaid balance with in-flight reservations: running pro/ultra tasks and whole Task Group batches reserve credit upfront, so a positive balance can still return 402. - Monthly spend limits are notify-only and never block requests. 16 moreShow fewer - Monitor bills per execution: an hourly lite monitor is ~720 executions a month, about $2.16, and 'core' monitor processor exists in the API with no published price. - FindAll fixed fee is charged per run even with few matches (pro $10 fixed); enrichments multiply Task price by matches; FindAll and Entity Search are public beta with 30 days notice of breaking changes. - FindAll rate limit conflicts: 300 per hour on the docs rate-limits page vs 25 / hr on parallel.ai/pricing. - /v1/extract moved full\_content, fetch\_policy, excerpt\_settings under advanced\_settings; sending v1beta top-level fields to /v1 returns 422. - Rate-limits page still names POST /v1alpha/monitors although Monitor is GA on /v1/monitors. - Customer Terms 2(b)/2(c)(vi): outputs are for the querying user / one end customer, may not be cached or shared across end customers, and may not be used to create databases, data brokerage or data selling products, or to train models. Matters for agencies storing results for reuse across clients. - Standard terms let Parallel train on inputs and outputs (de-identified); only enterprise ZDR and EU endpoint traffic are excluded. - AUP forbids building similar or competitive products and orchestrating usage across multiple accounts to dodge rate limits. - Public web only: no login-gated pages; Turbo search supports English and Japanese only. - The '-fast' Task processors (lite-fast, base-fast, core-fast, pro-fast, ultra-fast) announced Dec 2025 are no longer listed on today's processors or pricing pages; status not stated. - The Account API (balance, apps, keys) lives under `https://api.parallel.ai/account` and refuses API keys (401 'Missing bearer token'); it needs an OAuth access token. With a plain API key you cannot read your balance. - In-app balance and Usage round to the cent; Settings > Usage > EXPORT CSV gives exact per-call cost ($0.001 search fast, $0.001 extract, $0.005 task lite). - POST /account/service/v1/balance/add charges the org's default card immediately (amount\_cents + idempotency\_key); treat it as a paid call. - EU endpoint `https://eu.parallel.ai` (enterprise agreement only): Search requires mode 'basic' and Extract only; same x-api-key; about 1 hour to activate on a new key. - Search MCP caps excerpts at roughly 25,000 characters per tool call; anonymous calls run fast mode and ignore search overrides. - Latency bands differ by page: Task core is 60s-5min (docs pricing) vs 1-5min (pricing page/in-app); ultra8x 5min-2hr (docs) vs 8-30min (pricing page/in-app); lite 10-60s (docs) vs 5-60s (pricing page/in-app). ### Our take For an AI workforce startup client we ran the Task API on the base processor to find the top executive at each of 108 small companies: it returned a name for 76 and found nobody for 24. On a 181-contact research run for the same client, 14 Task runs did not finish within our polling window. We have also run pro-processor research reports for our own work. ## Who should use it Use Parallel AI for Row-level account/contact research and enrichment with per-field citations and confidence (Task API, $5-$2,400 per 1k runs) Look elsewhere for Pages behind a login (public web only, per FAQ) ## If Parallel AI is not the answer Other Research and scraping tools in this directory, each with its own review. [ ![](/_astro/exa.rOWjsHOa_ZIjcqx.webp) Exa Research and scraping Free planOfficial MCP Best for agent web search and RAG grounding with token-efficient contents. Exa API review →](/tool-directory/research-and-scraping/exa)[ ![](/_astro/dataforseo.lmwMYLks_Z1hF8Bv.webp) DataForSEO Research and scraping Best for cheap bulk SERP pulls via the Standard queue. DataForSEO API review →](/tool-directory/research-and-scraping/dataforseo)[ ![](/_astro/tavily.zLOQxnbo.svg) Tavily Research and scraping Free planOfficial MCP Best for agent web search with LLM-ready snippets at 1-2 credits per query. Tavily API review →](/tool-directory/research-and-scraping/tavily) ## Before you build Four things to settle first, each read off Parallel AI's own pages. - **Confirm the API is on your plan**The Free plan includes API access. Free tier: PUBLIC PAGES: 'Run up to 5,000 requests per month for free' and 'Earn up to $80 at signup + $5 in free credits per month'. - **Keep the key out of your code**API key. It is sent as: header x-api-key: for every product endpoint (spec ApiKeyAuth; the EU host eu.parallel.ai takes the same header). Account API (`https://api.parallel.ai/account/`...) needs Authorization: Bearer minted by Parallel OAuth (device / jwt-bearer grants); an API key gets 401 'Missing bearer token' (live test). MCP servers take Authorization: Bearer .. - **Design around the rate limit**Defaults per product, rolling 60-second window, only POSTs that create resources count (GETs and polling are free of limits). - **Know the catch before you scale**Search API default mode is 'advanced' at $5 per 1k; 'fast' is $1 per 1k and the docs call it the right choice for most agents. Set mode explicitly. ## Want us to build it for you? Tell us what you want Parallel AI to do in your stack. We will show you how we would wire it into your data, outbound or CRM, and what we would check first. [ Book a call ](/book)[ See case studies ](/case-studies) Read from [docs.parallel.ai/getting-started/pricing](https://docs.parallel.ai/getting-started/pricing), [docs.parallel.ai/api-reference/search/search](https://docs.parallel.ai/api-reference/search/search) and [docs.parallel.ai/getting-started/rate-limits](https://docs.parallel.ai/getting-started/rate-limits). Verified 25 Oct 2026. All 46 endpointsFor developers: method, path, what it does and what it costs. **Base URL:** `https://api.parallel.ai` **Bulk and async:** Task Group API: up to 1,000 runs per add-runs request, balance checked for the full batch upfront; Extract takes up to 20 URLs per call; FindAll runs up to 1,000 matches with exclude\_list up to 10,000; SQL/DataFrame integrations for BigQuery, Snowflake, Spark, DuckDB, Polars **Webhooks:** Yes: Task runs (event task\_run.status), FindAll runs and candidates, Monitor (monitor.event.detected, monitor.execution.completed, monitor.execution.failed). Standard Webhooks format with webhook-id, webhook-timestamp, webhook-signature (HMAC v1) using an account secret from Settings > Webhooks. Task webhook not available via the Python SDK. Filter endpoints 46 endpoints Method Path What it does Cost POST `/v1/search` Web search returning ranked URLs with LLM-ready compressed excerpts (GA) turbo/fast $1 per 1k requests; basic/advanced $5 per 1k requests; 10 results included, each additional result $1 per 1k; measured fast = $0.001. The response usage reports the same SKU (sku\_search) for fast and advanced. POST `/v1/images/search` Image search from keyword queries plus optional objective fast $1 per 1k requests; advanced $5 per 1k; up to 20 results per request at no extra cost POST `/v1/extract` Excerpts or full markdown content from given URLs (GA) $1 per 1k URLs ($0.001 per URL) per docs; in-app rate card and pricing page say '$1 per 1,000 results'; measured 1 URL = $0.001 POST `/v1beta/search` Legacy beta Search, kept only for existing integrations Not stated POST `/v1beta/extract` Legacy beta Extract, kept only for existing integrations; top-level excerpts/full\_content/fetch\_policy Not stated POSTasync `/v1/tasks/runs` Create a Task Run: web research or structured enrichment with Basis (citations, reasoning, confidence) Per successful run, by processor: lite $5/1k ($0.005), base $10/1k, core $25/1k, core2x $50/1k, pro $100/1k, ultra $300/1k, ultra2x $600/1k, ultra4x $1,200/1k, ultra8x $2,400/1k ($2.40/run). Failed runs not billed; measured lite = $0.005 GETasync `/v1/tasks/runs/{run_id}` Retrieve Task Run status Not stated GETasync `/v1/tasks/runs/{run_id}/result` Retrieve Task Run result, blocking until complete Not stated GETasync `/v1/tasks/runs/{run_id}/input` Retrieve Task Run input Not stated GETasync `/v1/tasks/runs/{run_id}/events` Stream Task Run progress events (SSE) Not stated POSTasync `/v1/tasks/groups` Create a Task Group to batch and track many runs Not stated POSTasync `/v1/tasks/groups/{taskgroup_id}/runs` Add runs to a Task Group (batch enrichment) Each run billed at its processor's Task price (see /v1/tasks/runs) GETasync `/v1/tasks/groups/{taskgroup_id}` Retrieve aggregated Task Group status Not stated GETasync `/v1/tasks/groups/{taskgroup_id}/runs` Fetch Task Group runs, optionally with inputs and outputs Not stated GETasync `/v1/tasks/groups/{taskgroup_id}/runs/{run_id}` Retrieve one Task Group run Not stated GETasync `/v1/tasks/groups/{taskgroup_id}/events` Stream Task Group events (SSE) Not stated POST `/v1/responses` Responses API: OpenAI Responses-compatible cited answers from live web research (replaces Chat in pricing tables) low $10/1k, medium $50/1k (default), high $250/1k successful requests POST `/v1beta/chat/completions` Chat API (beta): OpenAI ChatCompletions-compatible web-grounded chat; can also use Task processors $5 per 1,000 requests per product page [parallel.ai/products/chat](https://parallel.ai/products/chat); Chat is no longer listed on the docs pricing page or parallel.ai/pricing POST `/v1beta/findall/ingest` Turn a natural-language objective into a structured FindAll spec (entity\_type, match\_conditions) Not stated POSTasync `/v1beta/findall/runs` Create a FindAll run: discover, verify and optionally enrich all entities matching criteria (public beta) Fixed + per match: preview $0.10 + $0.00; base $0.25 + $0.03/match; core $2.00 + $0.15/match; pro $10.00 + $1.00/match. Enrichments billed per match at Task processor price GETasync `/v1beta/findall/runs/{findall_id}` Retrieve FindAll run status Not stated GETasync `/v1beta/findall/runs/{findall_id}/result` FindAll run result at time of request Not stated GETasync `/v1beta/findall/runs/{findall_id}/schema` Get FindAll run schema Not stated GETasync `/v1beta/findall/runs/{findall_id}/events` Stream FindAll events (SSE) Not stated POSTasync `/v1beta/findall/runs/{findall_id}/enrich` Add an enrichment (non-boolean fields) to a FindAll run Task API processor price x number of matches POSTasync `/v1beta/findall/runs/{findall_id}/extend` Raise match limit of an existing run Fixed cost not charged again; only per-match cost for new matches POSTasync `/v1beta/findall/runs/{findall_id}/cancel` Cancel a FindAll run Charged for work completed before cancellation POST `/v1beta/findall/entity-search` Entity Search (public beta): fast synchronous people or company search from a natural-language objective $5 per 1k requests incl. 100 results; additional results $0.05 per 1k POSTasync `/v1/monitors` Create a monitor (GA): event\_stream (query) or snapshot (task output) on a schedule Per execution: lite $3 per 1k, base $10 per 1k; core is in the API enum but has no published price GET `/v1/monitors` List monitors Not stated GET `/v1/monitors/stats` Aggregate monitor counts Not stated GET `/v1/monitors/{monitor_id}` Retrieve a monitor Not stated POST `/v1/monitors/{monitor_id}/update` Update a monitor Not stated POST `/v1/monitors/{monitor_id}/cancel` Cancel a monitor Not stated POSTasync `/v1/monitors/{monitor_id}/trigger` Trigger an immediate monitor run Not stated GET `/v1/monitors/{monitor_id}/events` List monitor events, newest first Not stated POST `/v1beta/memory/retrieve` Retrieve relevant or recent past Task/Monitor/FindAll runs from memory Not stated POST `/v1beta/memory/evict` Remove one run from memory Not stated POST `/v1beta/memory/clear` Clear a memory scope Not stated GET `/account/service/v1/balance` Account API: prepaid credit balance free [](/) © 2026 ONEGTMLAB Private Limited. All rights reserved. COMPANY [ Why Us ](/#why-us)[ Process ](/#timeline)[ Blogs ](/blogs)[ About ](/about)[ Clients ](/clients)[ Free resources ](/free-resources)[ Newsletter ](/newsletter) Services [ GTM Engineering ](/ai-gtm-engineering)[ Account-Based Marketing ](/services/account-based-marketing)[ Paid Media ](/services/paid-media)[ Content Engineering ](/linkedin-personal-branding)[ RevOps ](/services/revops)[ SEO, AEO & GEO ](/services/seo-aeo-geo)[ All services ](/services) ![](/_astro/wordmark-3d.CMEfGdYr.svg)