Locus product guide · Updated November 10, 2026

Deep research API options compared: cost per report

A deep research API takes one question, runs dozens of searches and page reads, and returns a cited report, usually in minutes. Google estimates $1 to $3 per Gemini Deep Research task and $3 to $7 for Deep Research Max. Parallel's Task API charges $0.10 to $2.40 per run on its research processors, Exa's Agent $0.012 to $1.00 per run, and OpenAI's o3-deep-research bills $10 per million input tokens and $40 per million output tokens, plus $10 per 1,000 web searches. Prices checked November 10, 2026.

What is a deep research API?

It is an agent behind one endpoint. It plans the research, searches, reads sources, searches again, and writes a report with citations. A search API returns links in a second. A deep research API returns an analysis in minutes.

Most run asynchronously. Gemini’s Deep Research agent requires background execution and allows up to 60 minutes per task. Parallel’s research processors take from 2 minutes to 2 hours.

How do the deep research APIs compare on price?

Billing models differ, which makes direct comparison hard. Some charge a flat fee per run. Others bill tokens and searches, so cost depends on how far the agent digs. Prices checked November 10, 2026.

How do the deep research APIs compare on price?
API	How it bills	Cost per report	Run time
Gemini Deep Research	Gemini tokens plus Google Search tool use	About $1 to $3, Google’s estimate	Minutes, 60 at most
Gemini Deep Research Max	Same, with more searches and tokens	About $3 to $7, Google’s estimate	Up to 60 minutes
OpenAI o3-deep-research	$10 per million input tokens, $40 per million output, plus $10 per 1,000 web searches	Varies with tokens and searches	Not listed
OpenAI o4-mini-deep-research	$2 per million input tokens, $8 per million output, plus searches	Lower, also variable	Not listed
Parallel Task API	Flat fee per successful run, by processor	$0.10 (pro), $0.30 (ultra), up to $2.40 (ultra8x)	2 minutes to 2 hours
Exa Agent	Per run, by effort	$0.012 to $1.00	Asynchronous
Tavily Research	Credits per request at $0.008 each pay as you go	$0.12 to $2.00 for pro, $0.032 to $0.88 for mini	Not listed
Perplexity Agent API, high effort	Model tokens plus $0.0025 per web search and $0.0005 per page fetch	Varies with model and tool calls	Not listed

Why are deep research costs hard to predict?

The agent decides how much to read. Google says a typical Gemini Deep Research task uses about 80 search queries, 250,000 input tokens and 60,000 output tokens. A Max task can reach 160 searches and 900,000 input tokens. A harder question costs more, and you learn the bill afterwards.

Flat per-run pricing removes that risk. Parallel charges one fee per successful run whatever the output, and does not bill failed runs. Exa and Tavily publish per-run bounds. Perplexity now recommends its Agent API for all new projects and maps its old Sonar Deep Research model to the Agent API’s high effort level.

Which research endpoints does Locus Pro offer?

Locus Pro offers grounded research endpoints with a price ceiling on every call, and the agent sees each price before it pays. Endpoints listed in the Locus catalog on November 10, 2026.

Locus Web Research searches the live web, reads sources and returns a cited answer. Its light effort level is priced up front. Parallel Responses runs Parallel’s web research in 5 to 60 seconds, priced by effort level. For multi-step goals, Locus tasks turn one plain-language goal into checked work with a ceiling of $5 by default and $200 at most, where tasks are enabled.

Some options are not available. The catalog excludes Perplexity Deep Research because its autonomous usage cannot be safely preauthorized, and it does not list OpenAI’s deep research models or the Gemini Deep Research agent.

Which research endpoints does Locus Pro offer?
Locus endpoint	How it is priced	What it returns
Locus Web Research, light effort	Priced up front	Cited answer with references
Parallel Responses	By effort level	Cited answer, optional JSON schema
Brave answers	Fixed price per call	Answer grounded in Brave search
Grok web search	Priced by request	Grok answer grounded in live search
Exa search, deep modes	Priced by request, with a cap	Results with content, from a seller Locus has not reviewed

What do 100 research reports a month cost?

Worked example, using list prices and Google’s estimates:

Gemini Deep Research: about $100 to $300.
Parallel Task API on the ultra processor: $30. On pro: $10.
Exa Agent: $1.20 to $100, depending on effort.

How do I run deep research from an agent without code?

Connect the agent to Locus Pro and ask the research question. Tell it: "Install Locus Pro by following https://paywithlocus.com/SKILL.md". Then ask: "Research the main competitors in EU battery recycling, cite sources, and spend at most 200 credits."

The agent checks prices for free, picks Locus Web Research or Parallel Responses, and passes your limit as the ceiling. If the live price exceeds the ceiling, the call is refused and nothing is charged.

Locus Pro for individuals

Which deep research API should I choose?

Pick by how long you can wait and how fixed the cost must be.

Long analyst-style reports in a Google stack: Gemini Deep Research.
Predictable cost per run at volume: Parallel’s Task API.
Cheap research rows in a spreadsheet or pipeline: Exa Agent or Parallel’s pro processor.
OpenAI models end to end: o3-deep-research or o4-mini-deep-research.
An agent that needs a cited answer in under a minute with a hard ceiling: Locus Web Research or Parallel Responses through Locus Pro.

Frequently asked questions

How long does a deep research API call take?

Minutes, usually. Google says most Gemini Deep Research tasks finish within 20 minutes, with a 60-minute maximum. Parallel’s research processors take 2 minutes to 2 hours. Parallel Responses answers in 5 to 60 seconds.

Can a deep research API return structured JSON?

Some can. Gemini’s Deep Research agent does not support structured output today. Parallel Responses accepts a JSON schema, and Locus Web Research accepts an output schema with an effort level.

Why can’t I run Perplexity Deep Research through Locus Pro?

The Locus catalog excludes it because its autonomous usage cannot be safely preauthorized. Perplexity itself now maps Sonar Deep Research to the high effort level of its Agent API.

Is a deep research API the same as a web search API?

No. A web search API returns ranked links in about a second. A deep research API runs many searches and reads, then writes a cited report, which takes minutes and costs more.

Implementation references

Use these first-party references for current request contracts and account requirements. Tool availability, prices, and negotiated terms can change.

Sources

Third-party details were checked on November 10, 2026. Vendors change pricing and features often, so confirm on their site before you decide.