AI agent cost calculator
Use this AI agent cost calculator to price conversations from steps, tokens, cache hits and extra billed calls. Add tools and seats, compare annual budgets, and copy the breakdown.
Free and without sign-up. The arithmetic runs in your browser, so nothing you type is sent anywhere.
Starting values are an AIGROW illustrative workload, dated 12 October 2026, not industry averages. Replace them with your own measurements. Every field is required; clearing a field stops the estimate.
$2 in and $10 out per million tokens, the price Anthropic publishes, read on 12 November 2026.
One model call each: a search, a read, a decision, a write. Capped at 40.
Count the tokens in instructions, available tools and the first request. Use provider usage logs or a tokenizer to measure them.
What the step brings back: the page it read, the rows it fetched, the result of the tool it called.
The answer only. Reasoning is the next field, because it is billed at the same rate.
Billed at the output rate. Enter 0 when no reasoning tokens are billed.
Use your measured cache-hit share. The first step is a cold read, so none of it is cached. Cache writes are excluded; enter their cost in other usage below.
Extra calls per original call: 10% adds one billed repeat per ten calls. This is not the probability of an unlimited retry loop.
How often the whole workflow runs: per ticket, per lead, per document.
- Read over 10 steps
- 107,500 tokens
- Of that, served from cache (70%)
- 72,450 tokens
- Written: the answer and the reasoning
- 9,000 tokens
- One run, 5% extra billed calls
- $0.1757
- 1,000 runs a month
- $175.71
List price as Anthropic publishes it, read on 12 November 2026. Cache hits at $0.10 per million. This is model usage only. Add hosting, tool calls, embeddings, logging and cache writes in the operating budget below. Long-context surcharges are excluded. Prices change; check the provider billing rules.
Formula: uncached input times input rate, cached input times cache-read rate, and answer plus reasoning times output rate, divided by one million; sum all steps and multiply by one plus extra calls and the chosen batch factor.
$175.71. What 1,000 runs of a 10-step agent cost a month on Claude Sonnet 5.5.
| Step | Reads | From cache | Writes | Cost |
|---|---|---|---|---|
| 1 | 4,000 | 0 | 900 | $0.0179 |
| 2 | 5,500 | 3,850 | 900 | $0.0133 |
| 3 | 7,000 | 4,900 | 900 | $0.0144 |
| 4 | 8,500 | 5,950 | 900 | $0.0154 |
| 5 | 10,000 | 7,000 | 900 | $0.0165 |
| 6 | 11,500 | 8,050 | 900 | $0.0175 |
| 7 | 13,000 | 9,100 | 900 | $0.0186 |
| 8 | 14,500 | 10,150 | 900 | $0.0197 |
| 9 | 16,000 | 11,200 | 900 | $0.0207 |
| 10 | 17,500 | 12,250 | 900 | $0.0218 |
Agent operating budget
$2.00 input, $10.00 output and $0.10 cache read per million tokens. Prices checked Nov 12, 2026. Provider price source Prices change. Confirm the current rate and billing rules before budgeting.
One conversation is one complete agent run. Model prices remain in USD; the manual comparison below uses the page currency and your exchange rate.
Optional. Blank excludes this cost; enter 0 to state zero.
Optional. Blank excludes this cost; enter 0 to state zero.
Optional. Blank excludes this cost; enter 0 to state zero.
Optional. Blank excludes this cost; enter 0 to state zero.
- Model cost per conversation
- $0.1757
- Operating cost per conversation
- $0.1757
- Fixed costs per month
- $0.00
- Yearly operating cost
- $2,108.55
Formula: model cost plus other usage per conversation, times monthly conversations, plus fixed costs and seats.
| Measure | Before | After |
|---|---|---|
| Conversations at 80% of your value | $175.71 | $140.57 |
| Conversations at 120% of your value | $175.71 | $210.85 |
| Billed tokens at 80% of your value | $175.71 | $140.57 |
| Billed tokens at 120% of your value | $175.71 | $210.85 |
AIGROW view setting, 12 October 2026: sensitivity changes volume or all billed tokens by 20%, holding model, token mix, retries and fixed costs constant.
Compare with manual handling
Use your own numbers. No wage, error rate or automation success rate is assumed. Amounts are in USD.
Monthly volume comes from the agent runs above.: 1,000
Required. Use your own measured or planned value.
Include pay, employer costs and overhead per productive hour.
Errors, rework, tools and seats
Optional. Blank excludes this cost; enter 0 to state zero.
Optional. Blank excludes this cost; enter 0 to state zero.
Optional. Blank excludes this cost; enter 0 to state zero.
Optional. Blank excludes this cost; enter 0 to state zero.
Optional. Blank excludes this cost; enter 0 to state zero.
Do not add software here if it is already included in your hourly overhead.
Formula: volume times handling and error rework minutes, divided by 60, times loaded hourly cost, plus tools and seats.
Enter valid volume, minutes and hourly cost. Large totals must remain within safe precision.
Set your automation scenario
Optional. Blank excludes this cost; enter 0 to state zero.
Costs and errors after automation
Optional. Blank keeps your current value; enter 0 to remove it.
Optional. Blank keeps your current value; enter 0 to remove it.
Optional. Blank keeps your current value; enter 0 to remove it.
Optional. Blank keeps your current value; enter 0 to remove it.
Optional. Blank keeps your current value; enter 0 to remove it.
Optional. Charged for every incoming item, including escalations. Exclude charges already counted in the model budget.
Optional. Blank excludes this cost; enter 0 to state zero.
The agent fixed costs above are included after automation: $0.00. Enter only additional fixed costs here to avoid counting them twice.
Enter your costs and expected share automated to compare. Clear or invalid inputs do not use a hidden estimate.
Released time is capacity. Cash savings require an actual reduction in spending.
Automation comparison is incomplete.
Sensitivity to volume and time
AIGROW view setting, 12 October 2026: a 20% scenario range. Change it to your own range; it is not a forecast.
All inputs stay in this browser. Blank optional costs exclude those costs from the estimate.
AIGROW illustrative workload, 12 October 2026: ten steps, 4,000 starting tokens, 70% cache hits after step one and 5% extra billed calls. On Claude Sonnet 5.5, reading 107,500 tokens costs $0.1757 per conversation, or $175.71 monthly for 1,000 conversations, before operating costs. Prices checked 12 November 2026. Eligible batch requests: 50% off.
How the number is worked out
Every figure the calculator assumes is on this page, so the result can be checked with a pencil.
- The prices
- Input, output and cache-read rates come from the repository model edition and link to the provider pricing page. Checked 12 November 2026. Prices change; confirm current billing rules. Model budgets stay in USD. EUR labour comparisons require your own exchange rate.
- The steps
- One conversation means one complete agent run. Each step reads the starting context plus what earlier steps added. Sum uncached input, cached input and output cost across all steps; the table includes the extra-call and selected batch factors.
- The cache
- The first step is a cold read. Later steps use your measured cache-hit share. With no published cache-read rate, cached tokens use full input price. Cache writes and long-context surcharges are excluded unless you enter their cost in other usage.
- Reasoning and retries
- Answer and billed reasoning tokens both use output price. Extra billed calls multiply the total by one plus the rate: 10% means one extra call per ten original calls. This does not model unlimited failure retries. Clearing a required field stops the estimate.
- The batch rate
- A batch discount is offered only when the selected model price notes publish one. Changing models resets the choice. No discount is applied when no batch rate is published.
- The operating budget
- Add hosting, seats, logging, embeddings and other tool usage once. Monthly cost is conversations times model and tool usage per conversation, plus fixed costs and seats. Yearly running cost is twelve months. Sensitivity changes conversations or billed tokens by 20%, an AIGROW view setting dated 12 October 2026, holding other inputs constant.
- Manual comparison
- Enter manual handling minutes and loaded hourly cost, then your expected share automated. Include review, errors and setup. Agent usage and fixed costs carry into the after-state. Released time is capacity; reducing cash spending requires a separate operational decision.
Questions people ask about this number
Price the agent on your own workflow
The audit measures the steps, context and volume of your workflow before anything is built.