Free calculator · per-resolution is the new pricing unit · prices verified 2026-09-19

Cost per task AI agent calculator: per ticket and per resolution

Cost per task for an AI agent is the unit the market now prices in, whether the task is a ticket, a claim or a document. Per-resolution vendors charge roughly $0.40 to $2.00; Intercom Fin at $0.99 is the reference. A custom agent costs cents in model tokens per conversation, about $0.004 for 3,700 tokens on Claude Haiku 4.5, plus a fixed monthly platform cost for hosting, observability and evals. Enter your ticket volume, tokens per ticket, model and resolution rate. You get cost per ticket handled, cost per ticket resolved, and monthly totals against the vendor and human paths.

15,000 tickets
3,700 tokens
55 %
$600 $
$9 $
$0.99 $
Custom agent cost per resolution
$0.102
$0.056 per ticket handled
Monthly cost, custom agent
$844
vs $8,168 at $0.99/resolution
Human cost avoided per month
$74,250
8,250 resolved x $9
Monthly cost by path at your volume (USD/month)
Custom agent
844
Per-resolution vendor
8.2K
Human cost avoided
74K
Monthly cost by path at your volume
Custom agent844.1999999999999
Per-resolution vendor8167.5
Human cost avoided74250
LLM cost per ticket (Claude Sonnet 5)
$0.016
Platform cost per ticket
$0.040
Tickets resolved per month
8,250
Benchmarks
Fin $0.99/resolution; eesel $0.40/ticket; Gorgias $0.90 to $1.00; Zendesk $1.50+; human $6 to $12
Anthropic worked example
~3,700 tokens per conversation, ~$37 per 10,000 tickets on Haiku 4.5
Prompt caching
not applied; see the caching calculator

A custom agent on Claude Sonnet 5 costs $0.102 per resolution against $0.99 from a vendor, and avoids $74,250 of human handling a month.

Illustrative model, not a quote. Cost bands are published third-party 2026 market ranges, never TheoSym or QGI prices. Per-token prices verified 2026-09-19; OpenAI prices are from a secondary tracker. Not legal, financial or tax advice.
Assumptions and sources (13)
ConstantValueBasis
Claude Haiku 4.5 price per MTok$1 in / $5 outsourcedAnthropic pricing page, fetched 2026-09-19
Claude Sonnet 5 price per MTok$2 in / $10 out (introductory price made permanent)sourcedAnthropic pricing page, fetched 2026-09-19
Claude Opus 5 price per MTok$5 in / $25 outsourcedAnthropic pricing page, fetched 2026-09-19
GPT-5.6 Luna price per MTok$0.2 in / $1.2 outsourcedSECONDARY: CloudZero GPT-5.6 pricing tracker (updated 2026-09-10). openai.com/api/pricing returned HTTP 403 on 2026-09-19
GPT-5.6 Terra price per MTok$2 in / $12 outsourcedSECONDARY: CloudZero GPT-5.6 pricing tracker (updated 2026-09-10). openai.com/api/pricing returned HTTP 403 on 2026-09-19
GPT-5.6 Sol price per MTok$5 in / $30 outsourcedSECONDARY: CloudZero GPT-5.6 pricing tracker (updated 2026-09-10). openai.com/api/pricing returned HTTP 403 on 2026-09-19
Gemini 3.6 Flash price per MTok$0.75 in / $3.75 out ($1.50 / $7.50 / $0.15 after 2026-12-31)sourcedGoogle Gemini API pricing page, fetched 2026-09-19
Token split70% input / 30% outputestimateStudio estimate; Anthropic's support example implies input-heavy conversations
Default tokens per ticket3,700sourcedAnthropic worked example (~$37 per 10,000 tickets on Haiku 4.5), 2026-09-19
Vendor reference price$0.99 per resolutionsourcedIntercom Fin ROI benchmarks
Vendor rangeeesel $0.40/ticket; Gorgias $0.90 to $1.00; Freshdesk $0.10 to $0.49/session; Zendesk $1.50+sourcedeesel comparison page
Human cost per ticket$6 to $12sourcedSame page
Default platform cost$600/monthproxyInside the $200 to $2,500 published run-rate band (SoftTeco)

Key takeaways

  • Model cost per ticket is cents. Platform cost per ticket falls with volume. Vendors charge dollars.
  • Cost per resolution, not cost per ticket, is the comparable number. Divide by what the agent finishes.
  • At low volume, per-resolution vendors win on speed. At high volume, custom wins on unit economics.
  • Resolution rate is the input that matters most. Measure it with evals before you trust it.

How cost per task for an AI agent is calculated

Tokens per ticket times the blended per-token price gives the model cost. Add the fixed monthly platform cost divided by tickets. That is cost per ticket handled. Divide by the resolution rate and you have cost per ticket resolved, the number that compares to a vendor quote.

Anthropic's own worked example is the anchor: about 3,700 tokens per support conversation, about $37 per 10,000 tickets on Haiku 4.5 (Anthropic pricing). Everything above that is platform, evals and the harness, which is what TheoSym's build packages actually price.

  • Cost per ticket = tokens x blended price + platform / volume
  • Cost per resolution = cost per ticket / resolution rate
  • Vendor monthly = resolved tickets x per-resolution price

AI vs human agent cost per ticket

Published 2026 benchmarks put a human-handled ticket at $6 to $12 and an AI resolution at $0.99 to $2.00 (eesel comparison). The gap is real but the comparison is often wrong: an agent that touches every ticket and resolves half leaves half the human cost in place.

This calculator multiplies your human cost by the share the agent actually resolves. The honest number is smaller and it survives the finance review. TheoSym's AI consulting starts by measuring that share with a 30-case eval, not by assuming it.

Build a support agent or pay per resolution?

At low volume, per-resolution pricing wins: no build, fast start, the vendor owns the harness. As volume grows the per-token economics take over, because model cost per ticket is cents. Published frameworks put the crossover near a million conversations a year (Digital Applied); your resolution rate and platform overhead move it.

The build path only wins if the agent ships with evals, an approval gate on refunds and account changes, and a kill switch. The AI Factory workproofs show that harness on shipped work. Regulated support flows, such as disputes and collections, route to QGI, where the decision layer has to be reproducible.

When do you need a cost-per-resolution model?

Before you renew a per-resolution contract, before you approve a build, and whenever the resolution rate changes. A vendor at $0.99 per resolution on 15,000 tickets and 55% resolution is about $8,200 a month. A custom agent at the same numbers is a few hundred dollars in tokens plus the platform line.

The saving is real. So is the work to earn it. Dr. Sam Sammane and the TheoSym team publish what shipped in the press center; bring your own numbers to the strategy call and leave with a scoped comparison.

cost per task AI agent: questions people ask

What is the cost per task for an AI agent, per ticket?+

Per-resolution vendors charge roughly $0.40 to $2.00, with Intercom's Fin at $0.99 as the reference price. A custom agent's raw LLM cost is far lower, around $0.004 per 3,700-token conversation on Claude Haiku 4.5, but you add hosting, observability and evals, which the calculator spreads across your volume.

What does a human-handled support ticket cost?+

Published 2026 benchmarks put a human-handled ticket at $6 to $12 depending on channel and complexity, versus $0.99 to $2.00 for an AI resolution. Enter your own fully-loaded figure; the calculator multiplies it by the share of tickets the agent actually resolves rather than assuming full deflection.

Is it cheaper to build a support agent or pay per resolution?+

At low volume, per-resolution pricing wins on speed and zero build cost. As volume grows, custom agents win on per-token economics because model cost per ticket is cents, not dollars. The crossover depends on your resolution rate and platform overhead, which is why both are inputs here.

Why does cost per resolution matter more than cost per ticket?+

An agent that touches every ticket but resolves half of them still leaves half the human cost in place. Dividing total agent cost by resolved tickets shows the true unit economics and makes vendor per-resolution pricing directly comparable to a build-your-own path.

Does the calculator include prompt caching?+

No. It prices every token at the base rate, which overstates the custom-agent cost for agents with a large stable system prompt. Cache reads cost 10% of base on most Claude models and 2.5% on Fable 5.1. Use the prompt caching calculator to see how much lower the model line goes.

Next step

Turn the estimate into a scoped plan

Send the calculator result. You get one reply from a human within one business day, with an eval plan for the workflow and a real scope.

Prefer to talk? +1 657-888-0688 or contact@theosym.com

Turn the estimate into a scoped plan

No spam. One reply from a human. Unsubscribe any time.