July 21, 2026

# Introducing the Parallel Responses API

Reading time: 3 min

Today we’re launching the Parallel Responses API[Parallel Responses API], the newest addition to our Web Agent API suite. The Responses API is an OpenAI-compatible endpoint for agentic web research, ranging from simple fact-retrieval-style questions to deep research reports. It turns any question into a synthesized answer, grounded in the live web, in seconds.

Illustration demonstrating deep research API concepts, web search capabilities, or AI agent integration features

## High quality web research in seconds

Like the Task API, the Responses API returns synthesized answers to research or enrichment tasks with full citations and support for structured outputs. While Tasks are designed for asynchronous or background work, providing the best cost to quality ratio, the Responses API is purpose-built for latency sensitive use cases. It provides the same quality as the Task API but significantly faster.

Responses	Task
low (5–10s, $10)	lite (10s – 60s, $5)
medium (15–20s, $50)	core (60s – 5min, $25)
high (30–60s, $250)	pro (2min – 10min, $100)

## Use Responses as a web research subagent

One key use case we had in mind while building Responses was subagents. Increasingly, agent builders are handing off well-scoped work to specialized subagents rather than stretching a single context window across everything. Web research is the natural work to hand off: it's token-heavy, it parallelizes well, and the orchestrator doesn't need to see how an answer was found — just the answer. The Responses API is that subagent, available to your agent via a tool call.

### Subagent

1
2
3
4
5
def web_research(query: str, effort: str = "low") -> dict:
   r = parallel.responses.create(
       model="parallel", input=query, reasoning={"effort": effort}
   )
   return {"answer": r.output_text}``` def web_research(query: str, effort: str = "low") -> dict:
   r = parallel.responses.create(
       model="parallel", input=query, reasoning={"effort": effort}
   )
   return {"answer": r.output_text}
```

Each `web_research` call is delegated to an autonomous researcher that runs its own multi-step searches, cross-checks sources, and returns a finished, cited answer — not raw results your agent has to read through. The division of labor pays off three ways:

- Context stays clean. Raw web results can run to tens of thousands of tokens on complex tasks. The orchestrator never sees them — each call returns just the answer, plus citations if you want them.
- Costs stay predictable. Orchestrators are usually the most capable models in the stack, which also makes them the most expensive place to process web content — every page pulled into context is billed at their rates, on every turn that follows. Research calls to the Responses API are a fixed price, however many pages the subagent reads.
- Research runs in parallel. Independent sub-questions fan out concurrently, so ten lookups take roughly the wall-clock time of one.

## OpenAI-compatible

The Responses API speaks the OpenAI Responses wire format. You can invoke Parallel’s Responses API using OpenAI’s Python or Typescript SDKs. Switching to Parallel is a 3-line diff:

The Responses API supports three reasoning tiers which, per the OpenAI standard, are set through the `reasoning.effort` parameter: `low`, `medium`, `high`.

Furthermore, the Parallel Responses API has full support for citations[citations], streaming via SSE[streaming via SSE], and stateful multi-turn conversations[stateful multi-turn conversations] through OpenAI’s response id parameter.

## Get started

The Responses API is available today.

Point your OpenAI SDK at `https://api.parallel.ai/v1`, set `model="parallel"`, and ask a research question.

### Give this to your agent:

1
Use curl to read parallel.ai/agents.md and perform the setup to install Parallel``` Use curl to read parallel.ai/agents.md and perform the setup to install Parallel
```

## Ready to get started?

Try Parallel[Try Parallel]Contact sales[Contact sales]

Are you an agent? Read this to onboard Parallel[Are you an agent? Read this to onboard Parallel]

By Parallel

July 21, 2026

## Related Posts93

Sep 28, 2026

- [Connect Lovable apps to the live web with Parallel](https://parallel.ai/blog/lovable-parallel)

Author: By Parallel

Sep 28, 2026

- [The outside world is the new infrastructure](https://parallel.ai/blog/the-outside-world-is-the-new-infrastructure)

Tags:Company

Author: By Liam Mulcahy

Sep 24, 2026

# Introducing the Parallel Responses API

## High quality web research in seconds

## Use Responses as a web research subagent

## OpenAI-compatible

## Get started

## Ready to get started?

## Related Posts93

- [Connect Lovable apps to the live web with Parallel](https://parallel.ai/blog/lovable-parallel)

- [The outside world is the new infrastructure](https://parallel.ai/blog/the-outside-world-is-the-new-infrastructure)

- [LangChain's Managed Deep Agents ships with built-in web search from Parallel](https://parallel.ai/blog/langchain-managed-deep-agents-parallel-search)

- [Bring trusted data into Parallel with Data Connectors](https://parallel.ai/blog/data-connectors)

- [Testing out Jev: real-world developer experience](https://parallel.ai/blog/testing-jev)

- [Introducing the Search Capability Leaderboard](https://parallel.ai/blog/search-leaderboard)

- [Merge launches Universal Context Layer with Parallel Search as its default web search provider](https://parallel.ai/blog/merge-universal-context-layer)

- [How to evaluate web search for AI](https://parallel.ai/blog/how-to-eval-web-search)

- [Building an apartment discovery tool with Parallel](https://parallel.ai/blog/apartment-finder)

- [Full Parallel API suite is now available on Google Cloud Marketplace](https://parallel.ai/blog/gcp-marketplace-full-api-suite)

- [Introducing Fast mode for Parallel Search](https://parallel.ai/blog/parallel-search-fast)

- [ Parallel ranks #1 on the Artificial Analysis Search Index, defining the Pareto frontier of quality, cost, and speed](https://parallel.ai/blog/artificial-analysis-best-search-api)

- [Always-on web monitoring research with Parallel Monitors and Hermes Agent](https://parallel.ai/blog/always-on-agents-monitor-api)

- [Building an always-on background agent to proactively support customers](https://parallel.ai/blog/customer-watch-background-agent)

- [Building a vendor intelligence system with Parallel](https://parallel.ai/blog/vendor-intelligence-system)

- [Parallel and Google Cloud Announce Partnership for Agentic Web Search on Gemini Enterprise Agent Platform](https://parallel.ai/blog/google-cloud-partnership)

- [$5 in free Parallel credits, every month](https://parallel.ai/blog/free-tier-parallel)

- [Introducing Turbo mode for Parallel Search](https://parallel.ai/blog/parallel-search-turbo)

- [Building a realtime voice agent with GPT-Realtime-2.1 and Parallel Search Turbo](https://parallel.ai/blog/gpt-realtime-parallel-turbo)

- [How Nooks cut web search costs 70.5% by switching to Parallel](https://parallel.ai/blog/case-study-nooks)

- [How Build created live geofenced alerts powered by Parallel for institutional real estate](https://parallel.ai/blog/case-study-build)

- [OpenClaw now has free, LLM-optimized web search by default powered by Parallel](https://parallel.ai/blog/free-web-search-openclaw)

- [Introducing real-time Entity Search](https://parallel.ai/blog/entity-search-company)

- [How we enrich & triage inbound leads using the Parallel Task API](https://parallel.ai/blog/enrich-triage-inbound-leads-parallel-task-api)

- [How AirOps creates citation-worthy content at scale, powered by Parallel](https://parallel.ai/blog/case-study-airops)

- [Introducing Index by Parallel](https://parallel.ai/blog/introducing-index-by-parallel)

- [Parallel Monitor API: New processor tiers, snapshots and event streams, and Basis on every event](https://parallel.ai/blog/monitor-api)

- [How we built parallelmpp.dev](https://parallel.ai/blog/parallel-mpp-dev)

- [How Actively's Per Account Agents use Parallel to turn the entire web into a proactive sales intelligence layer](https://parallel.ai/blog/case-study-actively)

- [Parallel Raises at $2 Billion Valuation to Scale Web Infrastructure for Agents](https://parallel.ai/blog/series-b)

- [Building a free CLI agent with Pi, Ollama, Gemma 4, and Parallel](https://parallel.ai/blog/free-CLI-agent)

- [Parallel Search is now free for agents via MCP](https://parallel.ai/blog/free-web-search-mcp)

- [Upgrades to the Parallel Search & Extract APIs](https://parallel.ai/blog/parallel-search-api)

- [How Finch is scaling plaintiff law with AI agents that research like associates](https://parallel.ai/blog/case-study-finch)

- [Genpact and Parallel Web Systems Partner to Drive Tangible Efficiency from AI Systems](https://parallel.ai/blog/genpact-parallel-partnership)

- [How Genpact helps top US insurers cut contents claims processing times in half with Parallel ](https://parallel.ai/blog/case-study-genpact)

- [A new deep research frontier on DeepSearchQA with the Task API Harness](https://parallel.ai/blog/deep-research)

- [How Modal saves tens of thousands annually by building in-house GTM pipelines with Parallel](https://parallel.ai/blog/case-study-modal)

- [How Opendoor uses Parallel as the enterprise grade web research layer powering its AI-native real estate operations](https://parallel.ai/blog/case-study-opendoor)

- [Introducing stateful web research agents with multi-turn conversations](https://parallel.ai/blog/task-api-interactions)

- [Parallel is live on Tempo, now available natively to agents with the Machine Payments Protocol](https://parallel.ai/blog/tempo-stripe-mpp)

- [How Parallel helped Kepler build AI that finance professionals can actually trust](https://parallel.ai/blog/case-study-kepler)

- [Introducing the Parallel CLI](https://parallel.ai/blog/parallel-cli)

- [How Profound helps brands win AI Search with high-quality web research and content creation powered by Parallel](https://parallel.ai/blog/case-study-profound)

- [How Harvey is expanding legal AI internationally with Parallel](https://parallel.ai/blog/case-study-harvey)

- [How Tabstack by Mozilla enables agents to navigate the web with Parallel’s best-in-class web search](https://parallel.ai/blog/case-study-tabstack)

- [Parallel Web Tools and Agents now available across Vercel AI Gateway, AI SDK, and Marketplace](https://parallel.ai/blog/vercel)

- [Authenticated page access for the Parallel Task API](https://parallel.ai/blog/authenticated-page-access)

- [Introducing structured outputs for the Monitor API](https://parallel.ai/blog/structured-outputs-monitor)

- [Introducing research models with Basis for the Parallel Chat API](https://parallel.ai/blog/research-models-chat)

- [Build a real-time fact checker with Parallel and Cerebras](https://parallel.ai/blog/cerebras-fact-checker)

- [Parallel Task API achieves state-of-the-art accuracy on DeepSearchQA](https://parallel.ai/blog/deepsearch-qa)

- [Introducing Granular Basis for the Task API](https://parallel.ai/blog/granular-basis-task-api)

- [How Amp’s coding agents build better software with Parallel Search](https://parallel.ai/blog/case-study-amp)

- [Latency improvements on the Parallel Task API ](https://parallel.ai/blog/task-api-latency)

- [Introducing Parallel Extract](https://parallel.ai/blog/introducing-parallel-extract)

- [Introducing Parallel FindAll](https://parallel.ai/blog/introducing-findall-api)

- [Introducing Parallel Monitor](https://parallel.ai/blog/monitor-api-beta)

- [Parallel raises $100M Series A to build web infrastructure for agents](https://parallel.ai/blog/series-a)

- [How Macroscope reduced code review false positives with Parallel](https://parallel.ai/blog/case-study-macroscope)

- [Introducing Parallel Search](https://parallel.ai/blog/parallel-search-api-beta)

- [Parallel processors set new price-performance standard on SealQA benchmark](https://parallel.ai/blog/benchmarks-task-api-sealqa)

- [Introducing LLMTEXT, an open source toolkit for the llms.txt standard](https://parallel.ai/blog/LLMTEXT-for-llmstxt)

- [How Starbridge powers public sector GTM with state-of-the-art web research](https://parallel.ai/blog/case-study-starbridge)

- [Building a market research platform with Parallel Deep Research](https://parallel.ai/blog/cookbook-market-research-platform-with-parallel)

- [How Lindy brings state-of-the-art web research to automation flows](https://parallel.ai/blog/case-study-lindy)

- [Introducing the Parallel Task MCP Server](https://parallel.ai/blog/parallel-task-mcp-server)

- [Introducing the Core2x Processor for improved compute control on the Task API](https://parallel.ai/blog/core2x-processor)

- [How Day AI merges private and public data for business intelligence](https://parallel.ai/blog/case-study-day-ai)

- [Full Basis framework for all Task API Processors](https://parallel.ai/blog/full-basis-framework-for-task-api)

- [Building a real-time streaming task manager with Parallel](https://parallel.ai/blog/cookbook-sse-task-manager-with-parallel)

- [How Gumloop built a new AI automation framework with web intelligence as a core node](https://parallel.ai/blog/case-study-gumloop)

- [Introducing the TypeScript SDK](https://parallel.ai/blog/typescript-sdk)