Skip to the page
Get started free

Perplexity deep research API in 2026

Cost, code and rows at 5¢.

Perplexity ended Sonar Chat Completions, home of sonar-deep-research, on September 27, 2026. Its replacement is the Agent API at high effort, billed in tokens plus tool fees such as $2.50 per 1,000 web searches, so the price is known after the run.

SpecStream is the live web data engine for teams and agents. Deep Search returns every match as a structured row at 5¢, 15¢ or $1, under a maximum you approve first. The Agent API can also call SpecStream as an MCP server.

Get an API key

New accounts get $10 of credit, no card.

SpecStream and Perplexity's deep research API side by side

SpecStreamPerplexity Agent API
The product todayDeep Search, with the Search API and the Extract API on the same keyThe Agent API, with effort levels fast, low, medium, high, xhigh and ultra. Sonar Chat Completions support ended on September 27, 2026
In place of Sonar Deep ResearchDeep Search: the request in plain words, run until it reaches the counteffort: "high": openai/gpt-6.1-terra with web_search and fetch_url, at most 40 tool calls. Perplexity names xhigh for its top deep research
What you sendThe data you want in plain words, with the conditions every row must meetA prompt as input and an effort level. You can override the model, max_steps and tools
What you get backOne structured row per match, with the columns you asked for and facts from many sites joined. Source URL and read time on every valueText with inline citations for source-backed claims, or JSON that follows a JSON Schema you pass in response_format
The unit you pay forA delivered row: 5¢, 15¢ or $1 by how much the row needs, plus 5¢ for an email and 15¢ for a phone. A run that finds nothing costs nothingTokens at each model's rates plus tool fees: web_search $2.50 per 1,000 calls, or $1.00 on Fast Search, fetch_url $0.0005 a call and sandbox $0.03 a session
The price before a runThe most the run can cost, quoted before it starts. Nothing runs until you approve it, and the charge never goes past itIt varies with the model, the tokens and the tool calls. usage.cost.total_cost reports the cost on a completed response
Long runsA run works until it reaches the count, up to four hours on the largest lists. The Search API answers in a few secondsBackground mode: submit to POST /v1/agent with background=true, then poll GET /v1/agent/{id}. Asynchronous Sonar requests are no longer supported
Tools and connectionsThe MCP server at https://api.specstream.ai/mcp, the REST API, Python and TypeScript SDKs, the CLI and A2ARemote MCP servers over Streamable HTTP with tool type mcp, connectors and your own functions

Why teams run their research on SpecStream

You get rows, ready for your table

Perplexity's high effort returns an answer with inline citations, and a schema shapes one response. Deep Search returns every match as its own row with the columns you asked for: companies, people, filings, tenders, papers or products. A result that fails one of your conditions is not delivered and not billed. get_find_decisions returns each condition's answer with its reason, the value, the source and the time it was read.

Source URL and read time on every value.

The price is set before the run

Perplexity's pricing page says effort level costs vary with the model, the tokens and the tool calls, and the cost is reported once the response is complete. On ultra, the model anthropic/claude-opus-5-5 costs $4.00 per 1M input tokens and $20.00 per 1M output tokens. SpecStream quotes the most the whole job can cost, good for 30 minutes, and the run starts only when you approve that maximum as confirm_max_cents. The charge never goes past it.

You pay for rows, and the searches behind them are free

On the Agent API, every web_search and fetch_url call is a fee on top of the tokens. On SpecStream a row costs 5¢ with up to 5 facts, 15¢ with up to 15 facts and one set of related records, and $1 for a deep row with no limit. The searches and pages behind each row are not billed, and a run that finds nothing costs nothing.

Keep the Agent API, and give it SpecStream's search

The Agent API connects remote MCP servers with a tool of type mcp: server_url, the key in authorization and allowed_tools. Tools you pass merge with the effort level's own. Add SpecStream at https://api.specstream.ai/mcp with search_web and read_pages, at $5 per 1,000 searches of up to 10 results and $0.80 per 1,000 pages. The code is in the next section.

The Search API on the same key is the most accurate there is

The SpecStream Search API scores 97.9 F-score on all 1,000 official SimpleQA Verified questions (978 correct), and 64.0 item F1 and 43.4 row F1 on all 200 official WideSearch tasks. The Extract API scores 0.908 F1 on WCXB. Scores and method are on the benchmarks page. The Search API costs from $1 per 1,000 searches of up to 10 results, and the Extract API costs $0.80 per 1,000 pages.

The rows stay up to date and land in your tools

Save the request and it reruns hourly to yearly. Each row comes back labeled changed, unchanged or no longer matching. A changed or new row costs half its rate and an unchanged row costs 1¢, so rechecking 1,000 rows that did not change costs $10. New matches arrive in Slack, email, a sheet, a CRM or a signed webhook.

Other deep research APIs

Gemini Deep Research

Two agents in preview on the Interactions API. Google estimates $1.00 to $3.00 a task, and $3.00 to $7.00 on Deep Research Max, at preview rates. A task runs at most 60 minutes, and the agent does not support structured outputs. SpecStream returns structured rows under a maximum you approve first.

Parallel Task API

Priced per run by processor, from $5 per 1,000 runs on lite to $2,400 per 1,000 on ultra8x. The ultra processor costs $300 per 1,000, or 30¢ a run, and takes 5 to 25 minutes. SpecStream bills the rows a run delivers, and a run that finds nothing costs nothing.

Tavily Research

One request takes 4 to 110 credits on mini and 15 to 250 on pro, at $0.008 a credit pay as you go, so 3¢ to $2 a request. The charge lands between the minimum and the maximum. SpecStream names the most a run can cost before it starts.

OpenAI deep research

OpenAI shut down o3-deep-research and o4-mini-deep-research on July 23, 2026, and its deprecations page names gpt-5.6-sol as the replacement. Web search there costs $10.00 per 1,000 calls plus search content tokens at model rates. SpecStream takes the same research request in plain words and returns rows at 5¢, 15¢ or $1 each.

How SpecStream scores

The same numbers as the benchmarks page, from the official tasks and scorers.

SpecStreamParallelExaTavilyPerplexity
Web researchWideSearch, GPT-5.6 Luna60.945.553.047.947.0
Factual accuracySimpleQA Verified, GPT-5.6 Luna9694919494
Factual accuracySimpleQA Verified, GPT-5.6 Sol9997919295

How to move a Perplexity deep research job to SpecStream

Write the data you want in plain words, in place of the research prompt you send today. Sign up and you get $10 of credit, no card. At 5¢ a row, that is up to 200 rows. Over MCP, your agent calls compile, quote_capacity_work and authorize_capacity_work, then reads the rows with list_find_rows.

Or send the request from code. This call spends nothing: the 402 reply shows how SpecStream read the request, the columns and the most it can cost.

curl -i https://api.specstream.ai/v1/agent/lists \
  -H "Content-Type: application/json" \
  -d '{"request": "AI startups that raised a seed or Series A round in 2026, with the founders, the round size, the lead investor and the product in one line", "count": 25}'

To keep the Agent API and add SpecStream's search to a high effort run, pass SpecStream as an mcp tool. Set SPECSTREAM_API_KEY to your SpecStream API key (MCP setup).

import os
from perplexity import Perplexity

client = Perplexity()

response = client.responses.create(
    effort="high",
    input="Compare the pricing pages of the five largest CRM vendors.",
    tools=[
        {
            "type": "mcp",
            "server_label": "specstream",
            "server_url": "https://api.specstream.ai/mcp",
            "authorization": os.environ["SPECSTREAM_API_KEY"],
            "allowed_tools": ["search_web", "read_pages"],
        }
    ],
)
print(response.output_text)
Get an API key

Frequently asked questions

Something we didn't answer? Talk to a human

Does Perplexity still have a deep research API?
Yes, as the Agent API. Sonar Chat Completions support ended on September 27, 2026, and Perplexity maps Sonar Deep Research to the high effort level, which it says often costs less per request. Asynchronous Sonar requests are no longer supported; long runs use background mode.
How much does the Perplexity deep research API cost?
Tokens at the model's rates plus tool fees: web_search costs $2.50 per 1,000 calls, or $1.00 on Fast Search, fetch_url $0.0005 a call and sandbox $0.03 a session. Perplexity gives no fixed price per run. Some third-party listings still show Sonar Deep Research at $2 per 1M input tokens and $8 per 1M output tokens. SpecStream bills 5¢, 15¢ or $1 a delivered row, under a maximum you approve first.
What replaced sonar-deep-research?
The Agent API with effort: "high", which runs openai/gpt-6.1-terra with web_search and fetch_url and makes at most 40 tool calls. Perplexity names xhigh for its top deep research. On SpecStream the same request in plain words returns structured rows at 5¢, 15¢ or $1 each.
Can the Perplexity Agent API use an MCP server?
Yes. Pass a tool of type mcp with server_label, server_url, the key in authorization and allowed_tools; the server must speak Streamable HTTP. SpecStream's MCP server is at https://api.specstream.ai/mcp, and its search_web tool costs $5 per 1,000 searches of up to 10 results.
Which deep research API returns structured rows?
SpecStream returns one structured row per match at 5¢, 15¢ or $1, with the columns you asked for and a source URL and read time on every value. A result that fails one of your conditions is not delivered and not billed.
Does SpecStream have an API and an MCP server?
Yes, and new accounts get $10 of credit, no card. The MCP server is at https://api.specstream.ai/mcp over Streamable HTTP, with OAuth or an API key. There is a REST API, Python and TypeScript SDKs, a CLI and A2A, and an agent can buy one request with no account, by card or in USDC.

Sources

SpecStream wrote this page. Each rival figure comes from that company's own page, and the third-party price from the listing named.

  1. SpecStream pricingspecstream.ai/pricing, read Nov 12, 2026
  2. SpecStream for agentsspecstream.ai/agents.md, read Nov 12, 2026
  3. SpecStream benchmarksspecstream.ai/benchmarks, read Nov 12, 2026
  4. Perplexity Sonar migrationdocs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview, read Nov 12, 2026
  5. Perplexity Agent API effort levelsdocs.perplexity.ai/docs/agent-api/effort, read Nov 12, 2026
  6. Perplexity pricingdocs.perplexity.ai/docs/getting-started/pricing, read Nov 12, 2026
  7. Perplexity Agent API modelsdocs.perplexity.ai/docs/agent-api/models, read Nov 12, 2026
  8. Perplexity background modedocs.perplexity.ai/docs/agent-api/background-mode, read Nov 12, 2026
  9. Perplexity output controldocs.perplexity.ai/docs/agent-api/output-control, read Nov 12, 2026
  10. Perplexity MCP tooldocs.perplexity.ai/docs/agent-api/tools/mcp, read Nov 12, 2026
  11. Puter: Sonar Deep Researchdeveloper.puter.com/ai/perplexity/sonar-deep-research/, read Nov 12, 2026
  12. Gemini Deep Research agentaistudio.google.com/docs/deep-research, read Nov 12, 2026
  13. Parallel pricingdocs.parallel.ai/getting-started/pricing.md, read Nov 12, 2026
  14. Tavily API creditsdocs.tavily.com/documentation/api-credits, read Nov 12, 2026
  15. OpenAI o3-deep-researchdevelopers.openai.com/api/docs/models/o3-deep-research, read Nov 12, 2026
  16. OpenAI o4-mini-deep-researchdevelopers.openai.com/api/docs/models/o4-mini-deep-research, read Nov 12, 2026
  17. OpenAI deprecationsdevelopers.openai.com/api/docs/deprecations, read Nov 12, 2026
  18. OpenAI pricingdevelopers.openai.com/api/docs/pricing, read Nov 12, 2026