Valyu DeepResearch or Hyperresearch: which deep research API fits the job

Use Valyu DeepResearch when you need research inside a product at a few cents to about $16 a task, structured output or Office files back, or sources Hyperresearch cannot read: licensed journals, market data, patents and the full SEC corpus behind one key. Use Hyperresearch when the report itself is the deliverable: $9 or $49 flat, built from 20 to 200 public sources kept in your workspace, with its cited claims checked against those sources before it ships.

Valyu sells two things developers compare with Hyperresearch: a Search API over the web plus specialised and proprietary indexes, and a DeepResearch API that turns one question into a cited report. This page covers both, with most of the weight on DeepResearch and its Heavy mode, using Valyu’s own docs and pricing as of October 2026.

Valyu DeepResearch and Hyperresearch side by side

Valyu’s cells come from its pricing page, DeepResearch guide, create and status API reference, data coverage and MCP docs, as of October 2026. Hyperresearch’s source counts, lengths and times are the targets the pipeline gates on, not guarantees.

Valyu prices a task, Hyperresearch prices a run

Both products charge a flat price per report, which makes them easy to compare. A Valyu task costs what its mode costs in credits: 8 for Instant, 25 for Fast, 100 for Standard, 500 for Heavy and 2,500 for Max, with the searches and page reads it performs included. Turning on optional tools adds 8 credits per screenshot, 16 per code run, 48 per browser session and 16 for each deliverable file after the first; charts and PDF reports are included. A completed task returns its total in cost with an itemised cost_breakdown.

One credit balance pays for Search, Contents and DeepResearch. Pay as you go buys 1,600 credits per $10 ($10 minimum top-up), so Standard costs about $0.63, Heavy about $3.13 and Max about $15.63. Monthly plans buy more credits for the money and unlock more sources: $29 a month for 5,000 credits, $99 for 18,000 and $499 for 100,000, with Enterprise priced on request. There are no overage charges: when the balance cannot cover a request, the API returns 402 Payment Required and does not run it. New accounts get $10 in free credits, or $20 with a company email.

Valyu’s DeepResearch guide gives each mode a hard time limit, 45 minutes for Heavy and 90 for Max, and recommends client timeouts of 60 and 120 minutes because a queued task can wait before it starts.

A Hyperresearch run costs $9 for Light or $49 for Deep. The price comes back in the same response that creates the run, before any work starts, and runs that fail on our side are refunded automatically. Your first Deep run is free once you add a card.

At Valyu’s pay-as-you-go rate, a Deep run costs about 16 Heavy tasks or a little over 3 Max tasks. That gap is real, and for many jobs it settles the question. The rest of this page is about what the $49 buys and when that matters.

Valyu reads licensed and proprietary data that Hyperresearch cannot

This is Valyu’s clearest advantage, and it is a big one for some readers. Its data coverage page describes a catalogue of papers and preprints, clinical and drug data, patents, market data, company financials, economic series, legal records, cybersecurity feeds and news, plus licensed Wiley journals and books. Some datasets are searched only inside a DeepResearch task. Which sources a task can reach depends on your plan: web, arXiv and PubMed are on every plan; the $29 plan adds market prices, bioRxiv and medRxiv; $99 adds SEC filings, clinical trials and economic data; and $499 adds patents, drug discovery, genomics and advanced company financials (pricing).

A Hyperresearch run reads what is public. It fetches web pages logged out, finds legal open-access copies of papers through Unpaywall, Europe PMC, PubMed Central and CORE, searches OpenAlex, PubMed and Crossref, reads SEC EDGAR filings, and pulls official statistics as tables. It does not read paywalled articles or paywalled papers with no open-access copy, and it has no connector or licence for private company databases. What a run can read lists every lane and what is out of reach.

So if your question turns on a Wiley journal article, intraday prices or a patent claim set, Valyu can reach evidence a Hyperresearch run cannot, and a Hyperresearch run will say in its notes that a source refused rather than guess at it. If you hold a licence to a document, you can upload it to your Hyperresearch workspace, where it stays in that workspace only.

Valyu returns sources, Hyperresearch returns a checked report

Valyu describes DeepResearch as an asynchronous research agent that searches multiple sources, analyses what it finds and writes a cited report. Each source in the result carries a title, URL, snippet, DOI where there is one, and often a fragment that deep-links to the cited passage on the page or in the PDF. That makes checking a citation by hand fast. What the status reference does not return is a per-citation result: there is no field saying whether each cited source supports the sentence that cites it. Checking that is your job.

A Hyperresearch run does part of that work before the report ships. Every cited figure is read against the source it cites, and what does not hold is corrected in one guarded revision. A Deep report adds a verification receipt with a verdict for each checked citation (supported, partially supported, unsupported or wrong source) and the passage from the source that the verdict rests on. A cited paper that has been retracted, or a quotation that does not appear in its source, stops the report from shipping. Syndicated copies of one article count as one voice.

A receipt is a record of checks, not a promise that every claim is true. A supported verdict means the source says what the sentence says, not that the source is right. Hyperresearch has not published a measured error rate for its own checker. A Light report gets the claim check but no receipt.

If you like Valyu’s sources and want the check, the two combine. The citation verification API takes a document and the sources it cites and returns a verdict per citation, a quote-integrity result and a retraction status per DOI. It works on documents Hyperresearch did not write, including another research API’s report. We have not tested it on Valyu output specifically.

On DRACO tasks, Hyperresearch scored 77.7 and Valyu DeepResearch Heavy 57.2

We graded Hyperresearch Deep and seven other research agents on the same 10 tasks from DRACO, a deep research benchmark from Perplexity and Harvard, in October 2026, and graded each with the same judge, GPT-5.2 at the paper’s settings. Valyu publishes its DeepResearch Heavy reports for DRACO in its valyu-benchmarks repository, so we graded those reports once each rather than running Valyu ourselves. Hyperresearch Deep runs were graded twice and the two grades averaged.

Valyu’s Shopping and General knowledge reports were not graded in this round, so its mean covers 8 tasks, from one grade each. Valyu scored higher on the Needle in a haystack task, 81.1 against 77.6, and came close on the UX design task, 72.6 against 74.2, and the Academic task, 87.8 against 91.2. The widest gaps are on the Medicine task, 72.2 against 0.0, and the Personalized assistant task, 64.1 against 26.4. DRACO subtracts points for advice its experts mark as harmful, and Valyu’s Medicine report took two of those penalties, one of them the largest on the rubric.

Three limits apply. Valyu’s published reports date from May 2026 or earlier, and today’s Heavy may score differently. These are 10 of DRACO’s 100 tasks, and two grades of the same report can differ by up to about 9 points. And this is our measurement, not an official leaderboard.

Valyu reports its own result: 72.7% for DeepResearch Heavy on all 100 DRACO tasks, one run per task, graded with Gemini 3 Pro as the judge. That number and ours cannot be compared. The judge and harness differ, and our 10 tasks are easier than DRACO’s average, so every system scores higher on them than on the full set. 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.

Where Valyu is the better choice

• Price per task. Heavy at about $3.13 and Standard at about $0.63, at the pay-as-you-go rate, put a cited report inside a product’s unit economics. Hyperresearch starts at $9.
• Speed. Instant answers in under 30 seconds, and Fast and Standard finish in minutes. A Hyperresearch Light run takes about half an hour, a Deep run hours.
• Licensed and proprietary data. Wiley journals, market data, patents, the full SEC filing set, genomics and more, on the right plan or licence. Hyperresearch reads public pages only.
• Output shapes. JSON to your own schema, PDF, and generated CSV, Excel, Word and PowerPoint files with charts. Hyperresearch returns a Markdown report with its evidence.
• Steering. Optional checkpoints pause a task to ask clarifying questions, review the plan, filter sources by domain or review the outline. You can attach up to 10 files and 10 URLs, connect up to 5 MCP servers as tools, and turn on code execution or browser sessions.
• Volume. A batch endpoint runs many tasks in parallel, and reusable templated Workflows (in beta) cover repeat jobs such as earnings previews.
• Raw search. If you are building your own agent loop, Valyu’s Search API prices retrieval per result: per thousand results, 100 credits for web, 50 for arXiv and PubMed, 1,920 for SEC filings and 4,800 for Wiley publications, or about $0.63, $0.31, $12 and $30 at the pay-as-you-go rate. Hyperresearch does not sell a search API.

Where Hyperresearch is the better choice

Hyperresearch fits when the report is the thing you hand over and someone will ask where each number came from. A Deep run splits the question into 3 to 5 research chapters, reads about 150 to 200 sources, maps where they contradict each other, sends investigators down the disputed points, and plans the report section by section before one writer writes it in one pass from all of the evidence, followed by the claim check and receipt. How it works lists every stage.

The second reason is what stays behind. Valyu lets a task take up to 3 earlier reports as context. A Hyperresearch run keeps every source it reads in your workspace vault with its provenance, searchable by words and meaning, and the next run in the same project searches the vault before the web. The vault exports as plain Markdown, so the corpus is yours.

The third is a fixed shape for agents. A Deep run takes hours, longer than any MCP client will hold a tool call open, so the Hyperresearch MCP server at https://mcp.hyperresearch.ai/mcp starts a run, returns its price and an id, and lets the agent check back. Long-running MCP tools explains the pattern. Valyu’s MCP server follows the same start-then-check pattern for DeepResearch.

The two calls side by side

A Valyu Heavy task:

curl -X POST https://api.valyu.ai/v1/deepresearch/tasks \
  -H "x-api-key: <VALYU_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "<question>",
    "mode": "heavy",
    "webhook_url": "https://example.com/hooks/valyu"
  }'

A Hyperresearch Deep run:

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

Moving a task across, field by field:

• query stays query, but Hyperresearch takes up to 4,000 characters against Valyu’s 25,000, so long briefs need trimming.
• mode becomes tier: light or premier, which is Deep on the wire. Omitted, it is light.
• research_strategy and report_format map to levers: register, purpose, reader_expertise, report_language, response_format, required_section_headings and domain_notes.
• search.included_sources has no equivalent. A run picks its own lanes from what a run can read.
• previous_reports becomes project_id. Runs filed under a project search its vault first.
• Polling GET /v1/deepresearch/tasks/{id}/status becomes GET /v1/runs/{id}, or use the webhook on either.
• The finished report comes from GET /v1/runs/{id}/result: report_markdown, sources, skipped, claims, verification, escalations and artifacts.

Both calls are asynchronous and answer 202. Valyu’s response carries the task id and its status; Hyperresearch’s carries the run id and price_cents before any work starts. A Hyperresearch run can also stop at blocked when a page needs a person, and you resolve it by uploading a copy you are entitled to. Pricing has the full numbers.

How this comparison was made

Every Valyu figure on this page comes from Valyu’s own docs, pricing page and benchmark repository, read logged out on October 12, 2026, with pricing, modes, time limits and rate limits re-read on October 22, 2026, and linked where it is used. Where Valyu does not publish a figure, such as sources per report or report length, the page says so instead of estimating. The quality numbers are our grading of Hyperresearch Deep runs and of Valyu’s published DRACO reports in October 2026, described on how we tested; ten tasks do not settle which writes the better report on your question. Hyperresearch figures are the targets the pipeline gates on, as printed on pricing. For other options, see the comparisons with Perplexity deep research, Gemini Deep Research and the OpenAI deep research API. We will re-check Valyu’s numbers by the end of 2026 and change the date at the top when we do.