Testing the full agent before paying anything
Pro · 8,000–20,000 credits
You want to see the real thing — the full agent, not the lightest one — on a job of your own before a card is charged.
$40/mo
$34/mo billed annually
The cheapest way to do that is not the cheapest rung. The free tier runs only the Lite agent, and the entry rung, per the Help Center, has no trial. The second card does: its rungs start with a seven-day free trial, which needs a card on file (Alipay is not accepted for trials). Cancel before the week is out and the Help Center says you are not charged; cancel a day late and the charge stands, because cancelling only stops the next renewal. Put the date in your calendar when you start.
A few serious tasks a week
Pro · 4,000 credits
You have a handful of research reports, decks or small sites to produce each week and no interest in running anything yourself.
$20/mo
$17/mo billed annually
The entry rung buys 4,000 credits a month plus 300 refresh credits a day, and the daily ones are worth less than they look: they only run the Lite agent, so anything on the full or Max agent comes out of the monthly 4,000. The vendor's own examples are the only yardstick available — a 15-minute data analysis at 200 credits, an 80-minute web app at 900 — which puts this rung at roughly twenty of the first or four of the second. Your tasks will not be the vendor's, and nothing on the page will tell you in advance how far off they are.
Working in it every day
Pro · 8,000–20,000 credits
It has become the place you start most pieces of work, several tasks run at once, and you ran out of credits in the second week.
$40/mo
$34/mo billed annually
Climb the second card before jumping to the third. Its four sizes — 8,000 to 20,000 credits — all sell at the same 200 credits per dollar as the entry rung, so moving up it changes how much you buy, not what a credit costs. Upgrading mid-cycle charges the price difference for the time left and issues the matching share of the extra credits, which makes it cheap to go up one size and see. The concurrency and scheduling limits do not change with the size: every paid rung allows 20 of each.
Batch work at volume
Pro · 40,000 credits and up
Wide Research across hundreds of items, recurring scheduled jobs, or a pipeline driving the API — tens of thousands of credits a month.
$200/mo
$167/mo billed annually
This is where the ladder finally bends. The 40,000 rung is the last one at 200 credits per dollar; from the next size up each dollar buys more, reaching 250 at the top of the list. It is also the tier the Help Center names as the only one allowed to buy add-on credit packs, which matters if your usage is spiky. One thing works in your favour here specifically: Wide Research is the single job for which the vendor publishes an upfront estimate and asks for confirmation, and it says it absorbs the cost when the estimate is exceeded.
A small team sharing one budget
Team
Three to ten people, uneven use — one heavy user, several occasional ones — and someone who has to answer for the invoice.
$20/seat/mo
$16.67/mo billed annually
Seats are priced on the personal ladder, but every seat's credits go into one pool, so the heavy user draws on what the light ones leave. Two differences are worth weighing before the pool. Team accounts get no daily refresh credits. And the data terms are better: the team page states that model providers are barred from training on Team and Enterprise data, and the Help Center lists opting out of data training as a Team feature — a line that does not appear on any personal plan. Below thirty seats, single sign-on is an extra flat fee.
Somebody has to sign it off first
Enterprise
A security and legal review sits between you and any rollout, and it will want contracts rather than a pricing page.
Contact sales
Go through sales. Team, the API and the vendor's other business services are governed by a separate Master Services Agreement rather than the consumer terms, and the consumer terms are the ones this page otherwise describes. Take three facts into the room: the vendor names no model behind the product, data is stored in the United States and Singapore, and the company changed owners twice between December 2025 and September 2026 — the public terms of use still predate both.