How many credits does a typical Agent Planners task use?

Measured on 337 production tasks: 349 at the median, 809 at the 75th percentile and 2,927 at the 90th. Plan pages budget 1,350–1,900, which 87% of tasks stay within.

What does a task cost in dollars?

Credits × your plan's price per credit — $0.001 on Starter. A median task (349 credits) is about $0.35; the plan pages' 1,350–1,900 budget is $1.35–$1.90.

Which tasks use the most credits?

Video creative, then SEO and web research and large dashboards: they call paid tools (a render, SERP pulls, a hosted browser) and read more data. Reports and single changes are the cheapest.

Do approvals cost credits?

No. Waiting for and giving an approval costs nothing; credits are spent by the model tokens and tools a task actually runs.

Billing

Credits by task type: what reports, changes, research and creative work cost

TL;DR

A typical Agent Planners task uses 349 credits at the median and 809 at the 75th percentile, measured on 337 production tasks. Research and video tasks use far more; this page breaks it down by task type and lists every tool's price.

Every task, measured

Over 90 days to 2026-10-02, 337 completed tasks used a median of 349 credits (about $0.35 at Starter's price), a 75th percentile of 809 and a 90th percentile of 2,927. The mean is 1,091 — a few long research and video tasks pull it up.

The plan pages budget 1,350–1,900 credits per task when they estimate how many tasks a plan covers. That is deliberately cautious: 82% of measured tasks used 1,350 credits or fewer and 87% used 1,900 or fewer, so you should not run out sooner than the plan page implies. Starter's 100,000 credits cover about 286 median tasks.

Model tokens are most of the bill; external tools — search, the sandbox, a browser, image or video — were 15% of measured spend. The model your agents run on moves the cost most: see credits by model.

By task type

Task type	What counts	Tasks	Median	≈ $ median	75th pct	90th pct
Proposes or makes a change	at least one write step — a budget, bid, status, campaign, audience or store change (approved by you by default)	248	409	$0.41	883	3,195
Autonomous run	the agent works in a loop, choosing its next tool itself, instead of following a fixed plan	85	408	$0.41	750	1,971
Single-platform report	reads and analysis on one platform, no change proposed	31	267	$0.27	1,362	2,631
Multi-platform analysis	two or more platforms in one task — e.g. Google Ads spend against GA4 or Shopify revenue	30	266	$0.27	596	5,489
General question	no platform data — an explanation, a plan or a draft	28	263	$0.26	1,234	1,785
SEO & web research	SERP / keyword research, web search, page scans or a hosted browser	19	1,496	$1.50	3,374	8,366
Dashboard build or refresh	a live dashboard built or refreshed by the agent	12	786	$0.79	8,061	9,968
Video ad creative	a video made by the creative studio or a generated clip	4	8,445	$8.45	21,690	21,690

Credits per task, measured on production tasks (model tokens re-priced at today's prices, tools at what they were billed). A task can be in more than one row — a multi-platform report that also proposed a change is in both — so the rows do not add up. Video has only a handful of runs; use its row as an order of magnitude, and the per-job estimate below to plan.

Tools: the price of one call

Tool	One call	Credits	≈ $ (Starter)
Web search	one search with cited results	67	$0.07
Google SERP lookup	one live search-results page	27	$0.03
Google Trends	one trends query	27	$0.03
Page scrape	one page read as clean text	67	$0.07
Page scan in a real browser	one hosted-browser visit	334	$0.33
Browser agent	one multi-step hosted-browser session	334	$0.33
Python sandbox	one analysis run (charts, joins, maths)	14	$0.01
Image generation	one image	267	$0.27
Email	one email sent	7	$0.01
MCP tool call (read)	one platform read by your own agent over MCP	7	$0.01
MCP tool call (write)	one platform write by your own agent over MCP	34	$0.03

The same on every model. SEO data (DataForSEO) and web data (Context.dev) are billed at the cost each call reports, so a cached SERP costs a fraction of a thousand-row keyword pull. Platform reads and writes made by the in-app agent have no per-call fee — only the model tokens that plan and analyse them.

Creative work

A generated video clip is priced by its length, resolution and model — a 5-second 720p clip on Veo 3.1 Fast is about 3,334 credits ($3.33).
A full video ad from the creative studio (a coding agent builds and renders it in a sandbox, with music) is estimated at about 27,334 credits ($27.33) before it starts, plus about 10,000 per extra platform rendition; you are billed what it actually measured when it finishes, and a workspace that cannot cover the estimate is told before anything runs.
An image costs 267 credits ($0.27).

How to spend fewer credits

Keep specialist agents on a cheap model (the default GPT-6 Luna is one of the cheapest) and reserve a strong model for the orchestrator.
Select only the accounts and date range the goal needs — fewer rows read means fewer tokens.
The Billing page shows the last 30 days of credits by model and by tool, so the expensive part is easy to find.

Frequently asked questions

How many credits does a typical Agent Planners task use?: Measured on 337 production tasks: 349 at the median, 809 at the 75th percentile and 2,927 at the 90th. Plan pages budget 1,350–1,900, which 87% of tasks stay within.
What does a task cost in dollars?: Credits × your plan's price per credit — $0.001 on Starter. A median task (349 credits) is about $0.35; the plan pages' 1,350–1,900 budget is $1.35–$1.90.
Which tasks use the most credits?: Video creative, then SEO and web research and large dashboards: they call paid tools (a render, SERP pulls, a hosted browser) and read more data. Reports and single changes are the cheapest.
Do approvals cost credits?: No. Waiting for and giving an approval costs nothing; credits are spent by the model tokens and tools a task actually runs.