Skip to content

Cost against price

What agents tools cost to run

20 tools in this category publish a monthly price. The median is $45, from $15 to $900. What follows is what the model underneath would cost at four shapes of work, and what that leaves for everything the company actually built.

One person using an assistant through a working day.

The median seat, $45, on the light workload

Cheapest priced model

Qwen 3.7 Flash

$0.13 to run

$44 left (100%)

Middle of the range

Gemini 3.1 Pro

$11 to run

$34 left (76%)

Dearest priced model

Claude Fable 5

$48 to run

$-3.50 left (-8%)

The same seat price, the same work, three model choices. Nothing about the product changed between those rows. That gap is the reason to ask a vendor which model is underneath and whether it can change without telling you.

Every agents tool that publishes a price

Entry price against the light workload on Gemini 3.1 Pro. Where the price is a range, the figure is the lowest published monthly number.

ToolEntry priceModel costLeft over
Felo$15$11$4.34 (29%)
Cognosys$15$11$4.44 (30%)
HARPA AI$15$11$4.44 (30%)
Lyzr$19$11$8.44 (44%)
Liner$20$11$9.44 (47%)
MindStudio$20$11$9.44 (47%)
Dust$29$11$18 (64%)
Flowise$35$11$24 (70%)
AgentGPT$40$11$29 (74%)
Factory$40$11$29 (74%)
Cassidy AI$49$11$38 (78%)
TaxGPT$50$11$39 (79%)
Dify$59$11$48 (82%)
Persana AI$85$11$74 (88%)
Gumloop$97$11$86 (89%)
Browserbase$99$11$88 (89%)
E2B$150$11$139 (93%)
Stack AI$199$11$188 (95%)
Warmly$700$11$689 (98%)
AiSDR$900$11$889 (99%)

A model, not an audit. We do not know which model any of these run on or how much work a seat does, and the figures exclude caching, batching and volume discounts, all of which push a real bill down.