ChatGPT deep research or Hyperresearch: which one fits the job

Use ChatGPT deep research when you already pay for ChatGPT and want a cited report in well under an hour, especially one that draws on your Google Drive, SharePoint or other connected apps. Use Hyperresearch when the report is the deliverable someone else will check: $9 or $49 flat, no subscription, with its cited claims read against their sources before it ships.

This page is for people using deep research in the ChatGPT app. If you were calling OpenAI’s deep research models from code, those were shut down on 2026-07-23, and migrating off the OpenAI deep research API covers what replaces them.

ChatGPT deep research and Hyperresearch side by side

ChatGPT includes deep research in every plan, metered as tasks. Hyperresearch sells one thing, a research run, through a web console, a REST API and an MCP server.

	ChatGPT deep research (app)	Hyperresearch
Price per report	Included in a plan. US prices: Free $0 and Go $8 a month (limited), Plus $20 (expanded), Pro from $100 (maximum), Business $20 a user a month billed annually	Light $9, Deep $49, flat, shown before the run starts. First Deep run free once you add a card
How usage is counted	Tasks, with an in-product counter. Counts per plan not published. Fixed monthly allowances reset every 30 days from first use	Per run. No allowance; a monthly spend cap you set
More reports past the limit	Business and Enterprise can buy credits, 50 credits a task. Credits on personal plans do not list deep research	Start another run
Typical time	“5 to 30 minutes” at launch in 2025; the current help page gives no figure	About 30 to 40 minutes for Light, about 3 to 5 hours for Deep
Sources per report	“Hundreds of online sources” at launch; not published now	About 20 to 50 for Light, about 150 to 200 for Deep
Report length	Not published	About 2,500 to 4,000 words for Light; Deep is sized to the question, up to about 9,000
Citation checking	Citations or source links on every report; no per-citation check result shown	Cited claims checked and corrected before shipping; Deep adds a verification receipt with a verdict per checked citation
Your own material	Uploaded files, chosen websites, and connected apps such as Google Drive and SharePoint	Uploads, kept in your workspace only. No Drive or SharePoint connector
Exports	Markdown, Word and PDF	The report in Markdown, with the full source list and claims table; the vault exports as plain Markdown
What is kept	The report stays in the conversation, under your retention settings	Every source a run read, in a searchable vault the next run searches first
API	None for app deep research	REST API with webhooks, and a remote MCP server

OpenAI’s cells come from its pricing page, its help article Deep research in ChatGPT, the ChatGPT rate card, the credits article for personal plans and the deep research launch post, all read on October 12, 2026. Hyperresearch’s source counts, lengths and times are the targets the pipeline gates on, not guarantees.

ChatGPT deep research limits are counted in tasks you cannot buy more of on a personal plan

OpenAI’s help article says “Deep research usage varies by plan. Your in-product usage counter shows your remaining tasks.” On plans with a fixed monthly allowance, the counter “resets every 30 days from the date of your first use”. The pricing page labels the allowance rather than counting it: “Limited deep research” on Free and Go, “Expanded” on Plus, “Maximum” on Pro. OpenAI published per-plan numbers when it raised the limits in April 2025; the current help page does not repeat them, so we do not either. Your counter is the only reliable figure.

Running out is where the plans split. On Business and Enterprise, OpenAI’s flexible pricing article says a user past their per-seat limit can keep going on credits the workspace has bought, and the rate card prices a deep research task at about 50 credits. If the workspace has no credits, “the feature is blocked”. On Free, Go, Plus and Pro, credits exist, but the personal-plan credits article lists Codex, ChatGPT Work, Word, Excel and PowerPoint as the supported features. Deep research is not on that list. A Plus user who runs out waits for the reset or upgrades.

That is the gap Hyperresearch fills for most people who arrive from ChatGPT. One report at a flat price, on the day you need it, with nothing to cancel afterwards. If you only hit the wall now and then, deep research limit reached lays out waiting, upgrading and buying one report side by side.

ChatGPT deep research cost per report depends on how much of your plan you use

Inside a plan there is no per-report price. If you pay $20 a month for Plus and run a few deep research tasks a week, each report costs you close to nothing beyond what you already pay, and you get the rest of ChatGPT with it. That is a better deal than any per-report purchase if research is a weekly habit.

On Business, the extra tasks past your seat’s allowance cost about 50 credits each on the rate card. The rate card does not print a dollar price per credit, so we cannot turn that into a figure.

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. You pay for the report you need and nothing in a month when you do not need one. Pricing has the billing and refund rules.

ChatGPT deep research for client work needs a citation check you do yourself

ChatGPT’s reports carry citations. OpenAI’s help article says “All deep research outputs include citations or source links so you can verify the information”, and the finished report has a sources-used section for that purpose. What you do not get is a result of anyone having checked them: we found no per-citation check in the help center.

OpenAI is direct about why the check matters. Its launch post says deep research “can sometimes hallucinate facts in responses or make incorrect inferences”, “may struggle with distinguishing authoritative information from rumors”, and shows weakness in confidence calibration. The launch post is from February 2025 and the product has changed since, but the current help article does not withdraw those limitations.

The one public number on ChatGPT’s citations comes from the DeepResearch Bench leaderboard. Its openai-deepresearch entry has a citation accuracy of 75.01% with 39.79 supported citations per task, as of October 2026. Put simply, about one citation in four did not hold up when the benchmark checked it. Its overall report score is 46.45, 29th of 45 entries on the main tab. The board’s second tab dates the OpenAI entry to 2025, so today’s product, on newer models, may score differently. Hyperresearch has no entry on that board.

If a client will read the report, one bad citation in four means someone reads every source before it goes out. A Hyperresearch run does that reading 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.

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 ten DRACO tasks, Hyperresearch scored 77.7 and ChatGPT deep research 55.2

We ran Hyperresearch Deep, ChatGPT 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 ChatGPT’s are each the mean of two grades.

System	DRACO tasks	Mean score
Hyperresearch Deep	10 of 10	77.7
ChatGPT Deep research †	10 of 10	55.2

† ChatGPT’s reports were exported from the app without source links, so its citation scores reflect that.

That footnote matters more for ChatGPT than for any other system on the board. DRACO has criteria that ask for named sources or links, and a report with no links scores zero on them. OpenAI’s help article says downloaded reports in Markdown, Word and PDF include citations or source links, so the copies we graded do not show everything the product can do on those criteria. Read the gap on citation-heavy tasks with that in mind.

The widest gaps are on the Finance task, 70.8 against 27.8, and the UX design task, 74.2 against 33.8. ChatGPT came closest on the Shopping task, 72.4 against 80.2, and the Needle in a haystack task, 68.7 against 77.6, and scored 81.3 against 90.6 on Law. 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.

Where ChatGPT deep research is the better choice

• Speed. OpenAI’s launch post put a task at 5 to 30 minutes. A Light run takes about 30 to 40 minutes and a Deep run 3 to 5 hours. If you need the answer before a meeting, use ChatGPT.
• Price for regular use. If you already pay for Plus or Pro and run research every week, the reports are effectively included. A single Light run is nearly half a month of Plus.
• Your connected apps. Deep research can read Google Drive, SharePoint, other connected apps and, according to OpenAI, authenticated industry data services available to your account, using read actions only. Hyperresearch reads public pages logged out and does not read anything behind a login or a paywall; your own documents reach it only as uploads.
• Control while it runs. You can review and edit the research plan before it starts, restrict or prioritise specific websites, follow progress and interrupt to change direction.
• The rest of ChatGPT. The report sits in a conversation you can keep asking questions in, and it downloads as Markdown, Word or PDF.
• Business administration. Business and Enterprise add admin controls, role-based access to deep research on Enterprise and Edu, and no training on business data by default. On personal plans, OpenAI’s pricing page says content is used to train its models unless you opt out.

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 lists every stage.

It fits when you do not want a subscription. Pay as you go has no monthly fee. A month with two reports costs the price of two reports, and a month with none costs nothing.

It fits when research compounds. Every source a run reads is kept in your workspace vault with its provenance, and the next run in the same project searches the vault before the web. A ChatGPT report lives in its conversation and is deleted with it. Hyperresearch keeps the source corpus itself, full-text searchable, and exports it as plain Markdown.

And it fits when an agent or a script starts the research. ChatGPT’s app deep research has no API. A Hyperresearch run starts with one REST call or one MCP tool call, returns its price and an id, and reports back by webhook or when polled. Long-running MCP tools explains the pattern. ChatGPT itself can reach the Hyperresearch MCP server as a developer-mode connector on the plans that allow it, though not as a deep research source; connecting Hyperresearch to ChatGPT says what works today.

Moving a ChatGPT deep research prompt to Hyperresearch

The prompt you wrote for ChatGPT works as it is, up to 4,000 characters. Three changes make the report sharper:

1. Name the decision. “Should we switch our 40 locations to a new point-of-sale system in 2027” gives a better report than “restaurant POS overview”.
2. Set the scope. Region, time period and the units you care about. If ChatGPT asked you clarifying questions, put your answers into the prompt.
3. Pick the size by breadth. One focused question is a Light. A question with several sub-questions, or one where you expect the sources to conflict, is a Deep.

If your ChatGPT report leaned on files from Drive or SharePoint, upload them to the run’s workspace. They stay in that workspace and are never reused across workspaces. What a run reads lists what it can and cannot reach.

How this comparison was made

Every OpenAI figure on this page comes from OpenAI’s own pricing page, help center and deep research launch post, or from the DeepResearch Bench leaderboard, read on October 12, 2026 and linked where it is used. Where OpenAI does not publish a figure, such as task counts per plan, report length or current run time, the page says so instead of estimating, and the two launch-era figures are dated as such. The quality numbers are the public leaderboard’s and our own grading of both products on 10 DRACO tasks 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. OpenAI has changed its plans, Pro tiers and credit rules within the last few months, so we will re-check these rows by the end of 2026 and change the date at the top when we do. For the other assistants, see Perplexity deep research and Gemini Deep Research.