Hyperresearch vs Gemini Deep Research

Choose Gemini Deep Research if you already pay for a Google AI plan, want a report in about 5 to 10 minutes, or need Gmail and Drive as sources. Choose Hyperresearch if someone will check the report: it reads more sources, checks cited claims against the pages they cite, and costs a flat $9 or $49 per report.

“Gemini Deep Research” now names two products. One is the Deep Research mode in the Gemini app, included in every Google AI plan. The other is the Gemini Deep Research agent in the Gemini API, a pay-per-use preview for developers. This page compares both with Hyperresearch, using Google’s own documentation as of October 2026.

The two Gemini products and Hyperresearch in one table

Sources for the Gemini columns: Use Deep Research in Gemini Apps, Gemini Apps limits, Google AI plans, Gemini Deep Research agent and Gemini API pricing, all read on 2026-10-01. Hyperresearch times and source counts are the targets the pipeline gates on, not guarantees.

Where Gemini Deep Research wins

Price, for most people. Deep Research appears on Google’s Free plan, and a Google AI Pro subscription at $19.99 a month gives 4x the Free plan’s usage along with everything else in the plan. If you run a few reports a week, the cost per report inside a subscription you already have is close to zero. A single Hyperresearch Light run costs $9.

Speed. Google’s help page says a report usually takes about 5 to 10 minutes. A Hyperresearch Light run takes about 30 to 40 minutes and a Deep run 3 to 5 hours. If you need the answer before a meeting that starts in a quarter of an hour, Gemini is the tool.

Google Workspace. The Gemini app can research across your Gmail and Drive, uploaded files and NotebookLM notebooks, and it exports the finished report to Google Docs. On Ultra, reports can include charts, diagrams and interactive simulators, though Google notes that visuals are not available when Workspace sources are included. Hyperresearch reads public pages logged out and has no Gmail or Drive connector; your own documents reach it only as uploads.

Planning and polish. Both Gemini products let you review and edit the research plan before work starts, and the app can turn a report into an Audio Overview.

Gemini Deep Research accuracy for client work

It is accurate enough for a first read, and for client work you should check its citations before you rely on them. Google says the same thing. Its developer documentation states: “We recommend reviewing the citations provided in the response to verify the sources.”

The only independent number on this comes from DeepResearch Bench, whose FACT evaluation checks whether each cited page supports the statement that cites it. Its Gemini entry is gemini-2.5-pro-deepresearch, the Gemini 2.5 Pro generation of Deep Research. On the board as of October 2026, that entry averaged about 211 citations per task, of which about 165 were judged supported: a citation accuracy of 78.3%. For comparison, the OpenAI deep research entry scored 75.0% with about 40 supported citations per task, so Gemini’s citation record was good relative to its peers. Still, roughly one citation in five did not hold up. The current Gemini app runs newer models than the benchmarked entry, so treat the figure as dated, not as a measurement of today’s product.

On the same board, the Gemini entry’s overall RACE score for report quality is 49.71; the top entry scores 58.03.

Hyperresearch handles the checking step inside the run instead of leaving it to the reader. A Light report’s cited claims are checked against the sources they cite and corrected where they do not hold before the report ships. A Deep report also ships a verification receipt: each checked citation carries a verdict of supported, partially supported, unsupported or wrong source, plus the passage from the source the checker relied on, so you can disagree with it. Unsupported or wrong-source citations fail the ship gate by default. A quotation that cannot be found in its source, or an unacknowledged citation to a retracted paper, blocks the report.

What the check does not do matters too. “Supported” means the cited source says what the sentence says, not that the source is right. Hyperresearch has not published a measured false-positive rate for its own checker. A receipt tells you what was checked and what was found; it does not replace judgment.

On ten DRACO tasks, Hyperresearch scored 77.7 and Gemini Deep Research 54.7

We ran Hyperresearch Deep, Gemini Deep Research and six other research agents on the same 10 tasks from DRACO, a deep research benchmark from Perplexity and Harvard, in October 2026. All were graded with the same judge, GPT-5.2 at the paper’s settings; our scores and Gemini’s are each the mean of two grades.

The widest gaps are on the UX design task, 74.2 against 21.6, and the Medicine task, 72.2 against 35.1. Gemini scored higher on the Needle in a haystack task, 78.6 against 77.6. Gemini’s Finance report came from the Gemini API and the rest from the Gemini app, copied out with their source titles but without links, so its citation scores reflect that. These are 10 of DRACO’s 100 tasks, two grades of the same report can differ by up to about 9 points, and this is our measurement, not an official leaderboard. How we tested has every task score and the other systems we graded, including Claude Research (Opus) at 73.0, or see a full Deep report.

Calling Gemini Deep Research from an API

Yes. The Gemini Deep Research agent is available through the Gemini API, with two limits to know before you build on it. It is in preview, and it is “exclusively available using the Interactions API”; you cannot reach it through generate_content. There are two agent ids, deep-research-preview-04-2026 for speed and deep-research-max-preview-04-2026 for maximum comprehensiveness. Every call must set background=true, and you poll or stream for the result.

By default the agent has Google Search, URL Context and Code Execution. You can add your own documents through File Search and connect remote MCP servers. Google lists three limitations: no custom function-calling tools, no structured outputs, and a maximum research time of 60 minutes. The agent documentation has the request shapes.

The Hyperresearch call is one POST:

curl -X POST https://api.hyperresearch.ai/v1/runs \
  -H "Authorization: Bearer <YOUR_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{ "query": "<question>", "tier": "light" }'

The 202 response carries the run id and price_cents before any work starts, so the price is known up front. Use "tier": "premier" for a Deep run. Poll GET /v1/runs/{id} or set webhook_url; the result returns the report, every source kept, every source tried and skipped with a reason, the claims table and, on Deep, the receipt. Agents can do the same over MCP; see connecting Claude and long-running MCP tools.

Because the Gemini agent accepts remote MCP servers, in principle you could point it at the Hyperresearch MCP server to search a vault you have already built. We have not tested that combination.

Gemini Deep Research cost per report

In the Gemini app, nothing beyond your plan. The US plans are Free at $0, Google AI Plus at $4.99 a month, Google AI Pro at $19.99 and Google AI Ultra at $99.99 or $199.99, as of October 2026. Usage is not counted per report. Gemini’s limits are compute-based, refresh “every 5 hours until you reach your weekly limit”, and Deep Research “will consume more of your usage” than ordinary prompts. The Deep Research help page also describes caps on daily and simultaneous research requests. Plus gives 2x the Free plan’s usage, Pro 4x, and Ultra 5x or 20x Pro. Without a plan, Google warns that Deep Research “may be unavailable during periods of high demand”. If you keep hitting the wall, deep research limit reached covers the options.

In the Gemini API, you pay for tokens and tools. Google’s own estimate, based on preview rates, is about $1 to $3 for a typical task (about 80 searches, 250,000 input tokens and 60,000 output tokens) and about $3 to $7 for a Max task (up to about 160 searches, 900,000 input and 80,000 output tokens). The pricing page bills model inference at standard Gemini rates, including reasoning tokens, and lists Grounding with Google Search for Gemini 3 models at 5,000 free requests a month, then $14 per 1,000.

Hyperresearch is $9 for Light and $49 for Deep, flat, whatever the run reads. A Gemini API task at Google’s estimate costs a fraction of a Light run. The extra buys a larger corpus, the claim check, the receipt on Deep and a vault that keeps every source. If you do not need those, the cheaper call is the right purchase. Pricing has the billing and refund rules.

Where Hyperresearch is the better choice

When the report leaves your hands. If a client, an investment committee or a reviewer will ask where a number came from, a Deep report gives you the source, the passage and the checker’s verdict for each checked citation, and refuses to ship on a fabricated quote.

When the question is broad and the evidence disagrees. A Deep run splits the question into 3 to 5 research chapters, reads about 150 to 200 sources, maps where they contradict each other and investigates the disputes before writing. How it works describes each stage.

When research compounds. Every source a run reads stays in your workspace vault, searchable by words and meaning, and the next run in the same project starts there instead of the open web. The vault exports as plain Markdown.

When you want a fixed price per report through an API, not a monthly plan or a token bill.

How this comparison was made

The Gemini rows come from Google’s own help pages, developer documentation, pricing page and plans page, read on 2026-10-01, and from the public DeepResearch Bench leaderboard data on the same day. We also graded Gemini and Hyperresearch on the same 10 DRACO tasks in October 2026, described on how we tested; ten tasks do not settle which writes the better report on a given question. Where Google does not publish a figure, such as sources per report or report length, the table says so instead of estimating. Hyperresearch figures come from its published pricing, and the times and source counts are pipeline targets. Gemini’s API agent is in preview, and its agent ids and prices are the parts most likely to change.