Deep research pricing: plans, APIs and the cost of one report

Deep research in a chat app costs whatever the plan costs: $0 on Gemini without a plan and on Perplexity Free, about $20 a month on ChatGPT Plus, Claude Pro or Google AI Pro, and up to $500 a month on ChatGPT Pro 500. None of the four apps (ChatGPT, Claude, Gemini, Perplexity) prices one report in dollars or publishes how many reports an individual plan includes, so there is no published cost per report. The APIs come closer. Parallel charges $0.30 to $2.40 a task on its deep research processors, Valyu about $0.63 to $15.63 a task at its pay-as-you-go rate, and Google estimates $1 to $7 a task for its Gemini Deep Research agent. OpenAI and Perplexity bill tokens and tool calls and publish no per-report price. Hyperresearch charges $9 for a Light run and $49 for a Deep run, flat.

Chat apps include deep research in a plan

Every figure below was read on the vendor’s own pricing or help page on October 22, 2026.

Plan	Price (US)	Deep research allowance
ChatGPT Plus	$20 a month	Included; count not published
ChatGPT Pro 100, 200, 500	$100, $200, $500 a month	More usage at each step; counts not published
ChatGPT Business, Standard seat	$25 a user a month, $20 billed annually	Per-seat limit, then about 50 credits a task from the workspace pool
Claude Free	$0	No Research
Claude Pro	$20 a month, $17 billed annually	Shares your chat limits, uses them faster
Claude Max	From $100 a month	Same, at 5x or 20x Pro’s usage
Gemini, no plan	$0	Included; may be unavailable at high demand
Google AI Plus, Pro, Ultra	$4.99, $19.99, $99.99 or $199.99 a month	2x Free usage on Plus, 4x on Pro, 5x or 20x Pro on Ultra; counts not published
Perplexity Free	$0	Limited
Perplexity Pro	$17 a month billed annually	Extended; count not published
Perplexity Max	$200 a month, $167 billed annually	Higher than Pro; count not published

Sources: ChatGPT Plus, ChatGPT Pro tiers, ChatGPT Business, ChatGPT rate card, Claude pricing, Research on Claude, Google AI plans, Gemini Apps limits, Perplexity pricing, Perplexity Max. ChatGPT also has Free and Go plans, but OpenAI’s plans page asked us to sign in on October 22, so their terms are not in the table.

The table has no cost-per-report column because none can be computed from public figures. A cost per report needs a report count and a price, and no vendor publishes both for an individual plan. ChatGPT Business prices a task at about 50 credits but prints no dollar price per credit. As a rough guide, a $20 plan costs as much as 2.2 Light runs at $9, so if its allowance covers three or more of your questions a month, the plan is cheaper.

Each product has its own comparison page: ChatGPT deep research, Claude Research, Gemini Deep Research and Perplexity deep research.

Deep research APIs bill per task, per token or per run

API	How it bills	Price per task
OpenAI (gpt-5.6-sol)	$4 per million input tokens, $20 per million output, $10 per 1,000 web searches	Not published
Perplexity Agent API	Model tokens, plus $0.0025 a web search and $0.0005 a URL fetch	Not published; $0.12 to $0.14 a request on Perplexity’s benchmarks (high)
Gemini Deep Research agent (preview)	Tokens and tools at standard Gemini rates	Google estimates $1 to $3, or $3 to $7 for the Max agent
Parallel Task API	Per completed run, by processor	$0.10 on pro; $0.30 to $2.40 on the four ultra processors
Valyu DeepResearch	Credits per task; 1,600 credits per $10 pay as you go	About $0.63 Standard, $3.13 Heavy, $15.63 Max; monthly plans cost less per credit
Hyperresearch	Per run	$9 Light, $49 Deep, returned before any work starts

Sources: OpenAI deprecations, OpenAI API pricing, Perplexity API pricing, Perplexity Agent API benchmarks, Gemini Deep Research agent, Parallel pricing, Valyu pricing, Hyperresearch pricing.

Some notes on reading the table. OpenAI shut down o3-deep-research and o4-mini-deep-research on July 23, 2026 and names gpt-5.6-sol as the replacement; its rate is promotional, available at least through November 21, 2026. Perplexity says Sonar, including Sonar Deep Research, is retired and maps it to the high preset; the xhigh preset cost $0.22 to $0.78 on the same benchmarks. Those are Perplexity’s measurements on benchmark questions, not a price. Google’s estimates are based on preview rates and, in its words, “subject to change”.

OpenAI’s replacement is a general model, so you build the research loop yourself. The other APIs run the loop for you. Migrating off the OpenAI deep research API, Parallel Task API and Valyu DeepResearch cover each one.

How deep research limits reset

• ChatGPT counts tasks on an in-product counter. On plans with a fixed monthly allowance it “resets every 30 days from the date of your first use”. On Business, a user past the seat’s limit draws on credits the workspace bought, and if there are none, “the feature is blocked”. ChatGPT Edu includes 5 deep research queries per rolling 24 hours, then uses credits.
• Claude draws Research from the same pool as your chats and coding. The pool resets on a rolling five-hour window, with weekly limits on top on paid plans, and paid plans can turn on usage credits to keep going.
• Gemini limits refresh “every 5 hours until you reach your weekly limit”, and Deep Research uses more of them than an ordinary prompt. Google also caps daily and simultaneous research requests.
• Perplexity resets Research on a rolling basis: each query leaves the usage window separately, so access comes back gradually. Renewing your subscription does not reset it.
• APIs have no allowance. They stop at rate limits (Parallel allows 2,000 task runs a minute by default) or at your balance (Valyu refuses a task its credit balance cannot cover).
• Hyperresearch has no allowance either. Each run is paid for on its own, and you can set a monthly spend cap.

Apart from ChatGPT Edu, none of the four apps publishes its deep research limit as a number. When you hit it, deep research limit reached sets out waiting, upgrading and buying a single report side by side.

The prices buy different things

ChatGPT reports come with “citations or source links so you can verify the information” and download as Markdown, Word or PDF. Perplexity exports Research to PDF, a document or a shareable Page. Gemini exports to Google Docs, and Google’s API docs recommend “reviewing the citations provided in the response to verify the sources”. In each case, checking the citations is left to the reader.

A Hyperresearch run does part of that check before the report ships. Every report arrives with the full list of sources read and a table of the claims it relies on, with the quote behind each. A Light report’s cited claims are checked against their sources and corrected where they do not hold. A Deep report adds a verification receipt recording which citations were read back against the source and what the checker found. Every source a run reads stays in your workspace vault, which the next run in the same project searches before the web. Light targets about 20 to 50 sources in about 30 to 40 minutes; Deep targets about 150 to 200 sources in 3 to 5 hours. Those are targets, not guarantees. On 10 DRACO tasks graded with the same judge, Hyperresearch Deep scored 77.7 and Claude Research (Opus) 73.0; how we tested has the method and its limits.

If you compare research APIs on price, divide by what you receive, not only by the call. Cost per verified citation sets out that unit and its method.

Which one to pay for

• You research every week inside a chat app you already pay for. Stay on the plan. Its cost per report falls with every report you run.
• You need research inside a product, at volume. Use a per-task API: Parallel or Valyu if you want the price fixed before the call, Perplexity or Gemini if token billing suits you.
• You need one report someone else will check, without a subscription. Buy a run: $9 for Light, $49 for Deep, priced before it starts. Runs that fail on our side are refunded automatically. Add a card and your first Deep run is free, one per new card and once per account. Pricing has the billing rules.