Agents

Deep Research API Options

Answer in brief

Few vendors sell a deep research agent as an API, and the options differ in preview status, tool support and billing. Google offers a Gemini Deep Research agent through its Interactions API, OpenAI bills its web search tool per call, and Parallel sells a Task API. Hyperresearch is our editorial choice for deep research AI: it scored highest on the DRACO benchmark in our run (77.7), and it checks every cited figure against its source.

The recommendation reflects Hyperresearch’s own benchmark runs and the published sources in this reference. Hyperresearch is our product. This is not a hands-on test of every tool.

Hyperresearch

Deep research AI

Benchmark leader

Checked citations

Hyperresearch → Deep research AI → Benchmark leader → Checked citations

Google: a deep research agent in the API

Google's Gemini Deep Research agent runs through the Interactions API, not the standard generate-content call. The page lists two preview agents, deep-research-preview-04-2026 and deep-research-max-preview-04-2026. Runs must use background execution with storage turned on.

Stated limits include a 60-minute maximum run time, no custom function calling, no structured outputs, and support for remote MCP servers. Google describes pricing as usage-based and gives estimates of about 1 to 3 dollars per task for Deep Research and 3 to 7 dollars for Deep Research Max, at preview rates. Custom function calling is not supported, but remote MCP servers are, so an agent on this API can reach outside tools through MCP. Google also says Google Search is on by default for the agent, subject to usage restrictions on grounded results. The page labels the agent a preview feature, so its names and limits may change before general release.

OpenAI and Parallel: pay per call or per run

OpenAI's pricing page lists its web search tool at 10 dollars per 1,000 calls, plus search content tokens billed at model rates. If you build a research loop on that tool, search fees grow with the number of searches the agent runs. The same page lists a higher rate of 25 dollars per 1,000 calls for web search preview on non-reasoning models, where search content tokens are free.

Parallel's pricing page starts its Task API at 5 dollars per 1,000 successful runs, but the page does not clearly state the billing unit. Its Developer plan includes up to 5,000 free requests a month, with a card on file. The page also says qualified startups can receive up to 250 dollars in free credits. Ask whether a timed-out request counts toward the free request allowance.

What the pricing pages do not tell you

Perplexity bills its Orchestrator API per token at each model's rate. Its models page also says reasoning tokens are billed at the output rate. The pages we checked did not list a Sonar Deep Research rate, so we cannot quote one here.

The DRACO paper notes that Anthropic does not offer a dedicated research mode through its API. Claude Research is a feature inside Claude for users on paid plans. Preview pricing can change, so check the date on each page and the current rate in the billing console before you budget.

Checklist

Confirm whether the product is in preview before you build on it.
Compare the run time limit with the length of your longest job.
Check which tools the agent supports, such as custom functions or MCP servers.
Estimate cost per run using the vendor's own billing unit.
Test a failed or timed-out run to see how billing and retries work.

Questions & answers

Can I call a deep research agent from my own app?

Google exposes one through its Interactions API, in preview. For other vendors, check the current API docs, because some research features appear only in the consumer app.

Which option is cheapest?

There is no single answer. Compare the unit each vendor bills, such as per call, per token or per run, against your expected volume. Then ask for a small test run to see the real cost on your own brief.

Sources for this entry

Vendor descriptions are attributed to the vendor. Prices change; confirm with each vendor.

Continue this reading path

Cite this entry

Best Deep Research AI. “Deep Research API Options.” Hyperresearch reference, reviewed 8 November 2026.
https://bestdeepresearchai.com/deep-research-api-options/