# Perplexity deep research or Hyperresearch: which one fits the job

Use Perplexity's Research mode when you want a cited answer in a few minutes and already pay for Pro. Use its Agent API when you need research inside a product at under a dollar a call. Use Hyperresearch when the report itself is the deliverable: $9 or $49 flat, with its cited claims checked against their sources before it ships.

If you came here looking for Sonar Deep Research, note that it changed on September 27, 2026. Perplexity's [migration page](https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview) says Sonar Chat Completions support ended that day: synchronous requests are being reformulated as Agent API requests, asynchronous Sonar requests are no longer supported, and the suggested replacement for `sonar-deep-research` is the Agent API `high` preset (formerly `advanced-deep-research`). Its [benchmark page](https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/benchmarks) now says Sonar is retired.

We make Hyperresearch, so read this as a vendor's comparison; the DRACO scores below are our own grading. It weighs cost per report, time, what each option can read, how citations are checked, and API and MCP access. Perplexity is the better pick on speed, price for quick questions, licensed data on Pro and Max, and API volume. If you are choosing between Perplexity and ChatGPT rather than a report service, see [Perplexity vs ChatGPT for research](/alternatives/perplexity-vs-chatgpt).

## The three options side by side

Perplexity sells deep research two ways: Research mode in its consumer app, and the Agent API for developers. Hyperresearch sells one thing, a research run, through a REST API, an MCP server and a web console.

| | Perplexity Research (app) | Perplexity Agent API, `high` preset | Hyperresearch |
|---|---|---|---|
| Price per report | Included in Pro at $17 a month billed annually, or Max at $167 a month billed annually ($200 a month in Perplexity's Max help article); limited on Free | Metered by tokens and tool calls; $0.12 to $0.14 per request on Perplexity's own benchmarks | Light $9, Deep $49, flat, fixed when the run is created |
| Typical time | "most research tasks in under 3 minutes" | "can take several minutes" | About 30 to 40 minutes for Light, about 3 to 5 hours for Deep |
| Sources per report | "dozens of searches, reads hundreds of sources" | Up to 15 steps, 15 results per search | About 20 to 50 for Light, about 150 to 200 for Deep |
| Report length | Not published | Up to 128,000 output tokens, a ceiling | About 2,500 to 4,000 words for Light; Deep is sized to the question, up to about 9,000 |
| Citation checking | Inline citations; no per-citation check result shown | Inline citations; no per-citation check result returned | Cited claims checked and corrected before shipping; Deep adds a verification receipt with a verdict per citation |
| API and MCP | App only (web, mobile, Mac); the API is billed separately | REST API with background mode; remote MCP server with a `perplexity_research` tool | REST API; remote MCP server with start, status and result tools |
| Workspace | Projects keep searches, files and instructions; up to 5 collaborators on Pro and Max | None built in; you carry context between calls | A vault that keeps every source a run read, searched first by the next run |
| Limits | Free limited, Pro extended, Max higher; counts not published; limits reset on a rolling basis | 1 to 33 requests per second, by spend tier | Spend cap per workspace; public pages and licensed news articles only, so no login-only sources |

Perplexity's cells come from its [pricing page](https://www.perplexity.ai/hub/pricing), [help center](https://www.perplexity.ai/help-center/en/articles/10738684-what-is-research-mode), [Max article](https://www.perplexity.ai/help-center/en/articles/11680686-perplexity-max), [presets](https://docs.perplexity.ai/docs/agent-api/presets), [MCP](https://docs.perplexity.ai/docs/getting-started/integrations/mcp-server) and [rate limit](https://docs.perplexity.ai/docs/admin/rate-limits-usage-tiers) docs, read on October 1, 2026, with prices, presets and Research limits re-read on October 22, 2026. Hyperresearch's source counts, lengths and times are the targets the pipeline gates on, not guarantees.

## Perplexity Sonar Deep Research cost per report

There is no per-report price. Perplexity bills its deep research by usage, so the cost is known after the call, not before it.

Perplexity no longer publishes `sonar-deep-research` rates. On October 1, 2026 its [pricing page](https://docs.perplexity.ai/docs/getting-started/pricing) still listed $2 per million input tokens, $8 per million output tokens, $2 per million citation tokens, $3 per million reasoning tokens, and $5 per 1,000 searches; by October 22 those rows were gone. The Agent API that replaces it bills tokens at the chosen model's rates plus a fee per tool call: $0.0025 per web search and $0.0005 per URL fetch. A completed response reports its own cost in `usage.cost.total_cost`.

For a sense of scale, Perplexity's [benchmark page](https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/benchmarks) lists cost per request on two public benchmarks, BrowseComp and DSQA. Sonar Deep Research, now retired, cost $0.57 to $0.61 a request. The `high` preset cost $0.12 to $0.14, and the `xhigh` preset (formerly `ultra`) $0.22 to $0.78. Those are Perplexity's measurements on benchmark questions, which ask for an answer rather than a long report, so treat them as an order of magnitude.

The [presets page](https://docs.perplexity.ai/docs/agent-api/presets) adds one thing to plan for: a preset called by name moves, and Perplexity can change the model behind `high` as its evals improve. Updates "target the same cost band", so the price should stay near where it is, but the output can change under you. You can pin a released version of a preset, or freeze its current values in your request, if that matters.

A Hyperresearch run costs $9 or $49, and the price comes back in the same response that creates the run, before any work starts. Runs that fail on our side are refunded automatically. That is roughly 66 to 405 times the `high` per-request figures above. The difference buys a report built from a target of about 20 to 200 sources and a claim check before release. Whether that is worth it depends on what happens to the report next.

## Perplexity deep research accuracy for client work

On citations, it scores well on the public benchmark. The [DeepResearch Bench leaderboard](https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard) lists `perplexity-Research` with a citation accuracy of 82.63% and 31.20 effective citations per task, as of October 2026. Put simply, about five in six of its citations held up when the benchmark checked them, which is among the higher citation scores on the board. Its overall report score is 40.46, 38th of 45 entries on the main tab.

Two caveats apply. The board does not say which version of Research it tested; its DeepResearch Bench II tab dates the Perplexity entry to 2025, and Perplexity has since shipped [Advanced Deep Research](https://www.perplexity.ai/help-center/en/articles/13600190-what-s-new-in-advanced-deep-research), so today's product may score differently. And Hyperresearch has no entry on the board, so the board says nothing about how the two compare; the DRACO section below is our own head-to-head.

For client work, the rate matters less than knowing which citations failed. If roughly one citation in six does not hold, someone has to find that one before the client does. Perplexity's answers carry inline citations, and checking them is your job: we found no per-citation check result in its API docs or help center.

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](/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.

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.

## On DRACO tasks, Hyperresearch scored 77.7 and Perplexity 57.8

We ran Hyperresearch Deep, Perplexity's deep research API (`sonar-deep-research`) and six other research agents on the same 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, mostly as the mean of two grades.

| System | DRACO tasks | Mean score |
|---|---:|---:|
| Hyperresearch Deep | 10 of 10 | 77.7 |
| Perplexity Deep Research (API) | 8 of 10 | 57.8 |

Perplexity's General knowledge and UX design reports were not graded in this round, so its mean covers 8 tasks. The widest gaps are on the Medicine task, 72.2 against 22.2, and the Personalized assistant task, 64.1 against 28.0. Perplexity scored higher on the Needle in a haystack task, 79.6 against 77.6, and came closest on the Academic task, 89.0 against 91.2, and the Law task, 87.5 against 90.6. 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](/benchmarks/draco) has every task score and the other systems we graded, including Claude Research (Opus) at 73.0, or [see a full Deep report](/samples/restaurant-pos-competitor-analysis).

## Where Perplexity is the better choice

- **Speed.** Minutes against half an hour or more. If someone is waiting on the answer, Perplexity wins.
- **Price for quick questions.** A Pro subscription gives extended Research use for $17 a month billed annually. A single $9 Light run costs more than half of that.
- **The app.** Research runs in the web, mobile and Mac apps, asks clarifying questions on broad queries, takes follow-ups while it works, and exports to PDF, a document or a shareable Page. Hyperresearch is a web console and an API.
- **Licensed data.** Perplexity's pricing page lists PitchBook, Statista and S&P Capital IQ as premium sources on Pro and Max. A Hyperresearch run reads public pages logged out and can buy a paywalled article under licence from 2,200+ news publishers, but it does not read private company databases or other paywalled articles.
- **Your own files and code.** Advanced Deep Research works with uploaded documents and runs calculations in a code sandbox.
- **Collaboration.** Projects can be shared with up to five people on Pro and Max.
- **API volume.** At under a dollar a request and up to 33 requests per second, the Agent API is built for research inside a product. Hyperresearch is not.

## 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, maps where its sources 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](/how-it-works) lists every stage.

The second reason is what stays behind. Every source a run reads is kept in your workspace [vault](/vault) with its provenance, and the next run in the same project searches the vault before the web. Perplexity's Projects keep conversations, files and instructions. Hyperresearch keeps the source corpus itself, full-text searchable, and exports it as plain Markdown.

The third is 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](/mcp/long-running-runs) explains the pattern.

## Moving a Sonar Deep Research integration to Hyperresearch

If each call needs a short cited answer, follow Perplexity's [migration guide](https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/how-to) to the `high` preset. It is the closest match and usually cheaper per request.

If you were using `sonar-deep-research` to produce long reports and then checking the citations by hand, the Hyperresearch call looks like this:

```bash
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"
  }'
```

`query` takes up to 20,000 characters. `tier` is `light` or `premier`, which is Deep on the wire. Every run is asynchronous: poll `GET /v1/runs/{id}` or wait for the webhook, then read the result for the report, the full source list, the claims table and, on a Deep run, the verification receipt. Your first Light report is free when you sign up, with no card. [Pricing](/pricing) has the full numbers.

## How this comparison was made

Every Perplexity figure on this page comes from Perplexity's own docs, help center and pricing page, or from the DeepResearch Bench leaderboard, read on October 1, 2026, with prices, presets, benchmark costs and Research limits re-read on October 22, 2026, and linked where it is used. The cost-per-request figures are Perplexity's own measurements, not ours. The quality numbers are the public leaderboard's and our own grading of both products on DRACO tasks in October 2026, described on [how we tested](/benchmarks/draco); 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](/pricing). Perplexity updates its presets in place, so we will re-check these numbers by the end of 2026 and change the date at the top when we do.
