AI Cost Tracker Calculator — Free LLM Cost Planning Tool

What Is the AI Cost Tracker Calculator?Free tier

Key User Pain Points — and How This Calculator Solves Them

Every indie developer and small AI product team hits the same wall: a token invoice arrives, and the number is far higher than anyone predicted. The calculator was built around seven specific failure modes.

Pain point	How the calculator solves it
"I can't estimate cost before I ship."	Model each feature's tokens, calls, and volume in the ledger, before writing a single API call.
"I don't know what prompt caching actually saves."	The Cache % column discounts cached input tokens to 10% of the normal rate, so you see the dollar impact immediately.
"My agent workflow cost 5× what I planned."	The Calls/action column multiplies cost per call by the number of LLM round-trips a single user action triggers.
"I don't know which model to use."	The Model swap comparison ranks your top features against four alternative models and shows the delta in dollars and percent.
"I found out I was over budget after the invoice."	The Budget guardrails panel projects month-end spend, days until budget hit, and daily burn rate.
"I don't know if my plan is profitable."	The Revenue & margin panel computes AI cost per customer and break-even customer count against your plan price.
"I don't know what to fix first."	The Optimization suggestions panel ranks fixes by dollar savings.

Units and Terminology

LLM pricing is quoted in US dollars per one million tokens. Every rate you enter follows that convention. Token counts are per call, not per month.

Field	Unit	Typical range	Notes
Input tokens	tokens/call	200 – 32,000	Everything you send: system prompt, context, user message
Output tokens	tokens/call	50 – 4,000	The model's reply
Calls per action	calls/action	1 – 20	1 for simple chat; 5–15 for agent workflows
Cache %	% of input	0 – 90	Fraction of input tokens served from a prompt cache
Batch?	on/off	—	Applies a 50% discount when the provider supports batch mode
User actions per month	actions/mo	100 – 10,000,000	Requests, generations, or workflows triggered by users
Input rate	USD / 1M tokens	0.075 – 15.00	Provider-published price for input
Output rate	USD / 1M tokens	0.30 – 75.00	Typically 3–5× input
Plan price	USD / month	5 – 200	What each paying customer pays you
Budget cap	USD / month	50 – 100,000	Maximum acceptable AI spend
Volume multiplier	multiplier	0.5 – 4	Applied to every row's request count
Day of month	day	1 – 31	Used to compute burn rate and month-end projection

Step-by-Step User Guide

Follow these twelve steps in order. The calculator recalculates every panel after each change — nothing needs to be saved manually, and your work persists across browser sessions.

1. Open the calculator and pick your segment.

   Use the Target customer dropdown to select your context: solo indie developer, small AI product team, agency, or internal team. This is a labeling convenience — it does not change the math.

2. Choose integration mode.

   Select Proxy, SDK wrapper, or Both. This documents how you plan to instrument cost tracking. It does not affect the calculation.

3. Set your currency (optional).

   Choose USD, EUR, GBP, INR, AUD, or CAD. When you pick a non-USD currency, an editable exchange-rate field appears. Every money figure in the calculator converts live.

4. Configure the volume multiplier.

   The Volume dropdown multiplies every row's request count by 0.5× (Low), 1× (Base), 2× (High), or 4× (Spike).

5. Add each AI feature to the ledger.

   Click + Add feature. Enter a descriptive name. Each row is independent — one feature per row. Rows are auto-numbered 01, 02, 03.

6. Pick a provider and model.

   Choose the provider from the dropdown (Anthropic, OpenAI, Google, Mistral, Other), then type the model name. Or use the toolbar's Preset picker to auto-fill the model name and both rates in one click. Presets apply to the last row you focused and never overwrite your token or volume entries.

7. Enter input and output tokens.

   These are per-call averages, not monthly totals.

8. Set calls per action.

   For a simple chat, leave this at 1. For an agent that does plan → search → summarize, enter 3 or 5. Every extra call multiplies the row's cost.

9. Enter cache % and batch mode.

   If your prompts repeat the same instructions across calls, both Anthropic and OpenAI offer prompt caching. Enter the fraction you expect to hit (0–90). Enable the Batch? checkbox only if a delay of hours is acceptable — it applies a 50% discount to the whole row.

10. Enter monthly request volume.

    How many times per month does one user trigger this feature? Multiply by the number of active users to get the row's Reqs/mo.

11. Enter the provider's published rates.

    Both In $/1M and Out $/1M come from the provider's pricing page. The presets give you editable starting points — verify against the source before finalizing.

12. Fill in revenue, budget, and read the outputs.

    Enter Plan price, Paying customers, and Other infra cost. Enter Monthly budget cap, Alert threshold %, and Current day of month. The sidebar, sensitivity table, chart, model swap comparison, and optimization suggestions all update instantly.

Complete Feature Reference — Every Field, Button, and Output

This section documents every element of the calculator. Use it as a map when you're looking for a specific control or want to understand what an output means.

The Ledger — Input Columns

The main table on the calculator is called the ledger. Each row represents one AI feature. Columns, in order:

Column	Input type	What it means	Default
#	auto	Row number, auto-assigned and renumbered when rows are removed	01, 02, 03…
Feature	text	Descriptive name for the feature ("Chat assistant", "Invoice parser")	empty
Provider	dropdown	Anthropic, OpenAI, Google, Mistral, or Other	Anthropic
Model	text	Model name as you'd write it in the API call	empty
In tok.	number	Input tokens per LLM call	0
Out tok.	number	Output tokens per LLM call	0
Calls/action	number	Number of LLM round-trips per user action (agents often use 3–5)	1
Cache %	number	Fraction of input tokens served from a prompt cache	0
Batch?	checkbox	Applies a 50% discount when enabled	unchecked
Reqs/mo	number	User actions per month for this feature	0
In $/1M	number	Input rate in USD per one million tokens	0
Out $/1M	number	Output rate in USD per one million tokens	0
Cost/req	auto	Computed cost per individual LLM call	—
Cost/mo	auto	Computed monthly cost for this row (after calls/action and volume multiplier)	—
×	button	Removes the row	—

The footer row of the ledger shows Total monthly AI cost — the sum of every row's Cost/mo.

Free-tier cap

Free users can add up to 3 rows. The + Add feature button disables at that point, and a note appears below the button explaining the limit.

The Toolbar — Controls

Above the ledger sits the toolbar. Left to right:

Control	What it does
Volume dropdown	Multiplies every row's requests by 0.5×, 1×, 2×, or 4×. Base is 1×. Choose High or Spike to stress-test the plan under heavier traffic without retyping numbers. A custom value persists if you load a saved plan that used one.
Preset picker	Dropdown grouped by provider. Selecting a model fills the last-focused row's provider, model, in-rate, and out-rate. Never touches your tokens, calls per action, cache, or volume.
Export JSON	Downloads the entire plan (all scenarios, all rows, currency, revenue, budget, notes) as a `.json` file. Use for backup or to move the plan to another browser.
Import JSON	Reads a previously-exported file and replaces the current plan. Respects your tier's scenario limit.
Print / PDF	Opens the browser print dialog with a print-specific layout: controls hidden, inputs flattened, summary preserved. Save as PDF or send to a physical printer.
Copy CSV	Copies the current ledger as comma-separated values to your clipboard. Suitable for pasting into Excel, Google Sheets, or Numbers. Includes all columns plus a TOTAL row.

Scenario Tabs

Below the toolbar sits a row of tabs. Each tab is an independent ledger with its own rows. Use scenarios to model different pricing tiers (Free/Pro/Team), different models, or different growth assumptions side by side.

Action	How
Add a scenario	Click the + tab at the end of the row. Free tier: 1 scenario total; Pro: 3; Team: 6. When you copy a scenario, it inherits the current rows as a starting point.
Switch scenario	Click any tab. The ledger and all outputs refresh to reflect that scenario's rows.
Rename scenario	Double-click the tab. A prompt asks for the new name (max 40 characters).
Close scenario	Click the × on a tab. At least one scenario must remain.
Current scenario	Highlighted in a lighter background. Its name also appears next to the "Feature cost ledger" heading.

Revenue & Margin Panel

Below the ledger, this section ties cost to what customers pay.

Inputs in this panel: Plan price, Paying customers, Other infra cost. Outputs:

• Monthly revenue = plan price × paying customers
• Total variable cost = total AI cost + other infra cost
• Gross margin = margin percentage, or — when revenue is zero
• AI cost per customer = total monthly AI cost ÷ paying customers
• Break-even customers = ⌈ variable cost ÷ plan price ⌉

Budget Guardrails Panel

Sets a spending ceiling and projects where you'll land.

Field	What it does
Monthly budget cap	The maximum acceptable AI spend. Drives the budget status pill and the burn bar in the sidebar.
Alert threshold %	When total cost crosses this fraction of the cap, the sidebar shows "near threshold" in amber. Default 80%.
Notification channel	Documentation field. Email, Slack, Webhook, or In-app only. Recorded in the plan but doesn't trigger real notifications from a planning tool.
Enforcement mode	Documentation field. Alert only, Block requests, Route to fallback model, or Require approval. Recorded for use when you build the real system.
Current day of month	The day the plan was last updated. Used to compute daily burn rate and month-end projection.

Outputs:

• Month-end projection = burn rate per day × 30
• Days until budget hit = remaining budget ÷ daily burn rate
• Burn rate = total monthly cost ÷ day of month

Sensitivity Panel

A heat-mapped grid showing total cost across combinations of request volume and output-token length.

Axis	Values	Meaning
Columns (Reqs ×)	0.5, 1, 2, 4	Request multiplier applied to every row
Rows (Out tok ×)	0.5, 1, 2	Output-token multiplier applied to every row
Cell color	Green / Amber / Red	Relative to the grid's own minimum and maximum, not the budget cap. Green = cheapest cell; Red = most expensive.

Cost by Feature Chart

A horizontal bar chart, sorted descending by monthly cost. Up to 8 features shown. Bars scale to the highest-cost row, so you can see relative weight at a glance. Values are printed next to each bar in compact form (e.g. $76.80, $1.2k).

Model Swap Comparison

Compares your top three most expensive features against four alternative models:

• Claude Sonnet ($3 / $15 per 1M)
• Claude Haiku ($0.80 / $4 per 1M)
• GPT-4.1 mini ($0.40 / $1.60 per 1M)
• Gemini 1.5 Flash ($0.075 / $0.30 per 1M)

Each cell shows the alternative monthly cost, the dollar delta, and the percentage change. Green cells mean cheaper; red cells mean more expensive. The comparison uses your exact row settings (tokens, calls per action, cache %, batch), and only swaps in the alternative model's rates.

Optimization Suggestions Panel

An automatic list of ways to reduce monthly spend, ranked by dollar savings. Up to five suggestions shown. Each suggestion covers a specific row and quantifies the saving. Categories:

1. Cheaper same-provider model — e.g. Claude Sonnet → Claude Haiku on one row
2. Enable prompt caching — assumes 80% cache hit rate for the calculation
3. Move to Batch API — assumes 50% discount
4. Reduce output tokens by ~50% — suggests tighter prompts or max_tokens caps
5. Consolidate multi-call workflows — suggests dropping one round-trip per action

The total of all suggestions appears in the sidebar as Possible savings.

Assumptions & Warnings

Two stacked blocks. Warnings lists automatic flags raised by your current inputs. Assumptions lists the planning assumptions with confidence chips.

Warning conditions:

• Rate above $500 per 1M tokens on any row
• Token count above 500,000 per call
• Request volume above 10M per month
• Row with zero requests but non-zero tokens
• Multi-call workflow combined with high cache %
• AI cost per customer exceeding plan price (red error)
• Negative gross margin (red error) or below 20% (amber warning)

Assumption rows include total requests/month, total LLM calls/month, total input and output tokens/month, rows on Batch API, average cache hit rate, providers in use, active scenario, volume multiplier, currency and rate, and how rates were sourced. Each row shows a high, medium, or low confidence chip.

Summary

The right-hand panel (or stacked below on mobile) shows at-a-glance totals:

Below the readouts:

• Costliest features — top 5 by monthly cost with percentage of total
• Budget burn bar — visual fill showing percentage of cap used, color-coded green/amber/red
• Copy full plan button — premium (Pro/Team) feature. Copies a plain-text summary including all formulas, per-feature lines, revenue and margin, budget guardrails, and optimization suggestions.
• Clear ledger button — resets the active scenario to default rows after a confirmation prompt

Scenarios, Currency, and Persistence

Scenarios: Each tab in the scenario bar is an independent ledger. Add up to 1 (Free), 3 (Pro), or 6 (Team). Rename by double-clicking a tab. Close with the × icon. Switching tabs swaps the entire ledger and all outputs.

Currency: Pick from USD, EUR, GBP, INR, AUD, or CAD. All money readouts convert using a fixed default rate table (EUR at 0.92, GBP at 0.79, INR at 83.5, AUD at 1.52, CAD at 1.36). Non-USD options reveal an editable rate field so you can match your bank's rate. The rate you set is saved with the plan.

Persistence: Every change is saved automatically to your browser's localStorage. Closing the tab and reopening restores the plan exactly. Clearing browser data for the site will wipe the plan — use Export JSON for a portable backup. Use Import JSON to restore or transfer the plan to another device.

Shareable links: The calculator exposes a helper that encodes the plan into the URL hash (#plan=...). Opening a link with that hash loads the encoded plan. Anyone with the link sees the same state, which is useful for sharing a scenario with a teammate or saving a snapshot to bookmarks.

Design Preset Rates (Editable Starting Points)

The preset picker uses the following rates. They are starting points, not live quotes. Edit them in the ledger after applying, or verify against the provider's current pricing page.

Provider	Model	In $/1M	Out $/1M
Anthropic	Claude Opus	15.00	75.00
Anthropic	Claude Sonnet	3.00	15.00
Anthropic	Claude Haiku	0.80	4.00
OpenAI	GPT-4o	2.50	10.00
OpenAI	GPT-4o mini	0.15	0.60
OpenAI	GPT-4.1 mini	0.40	1.60
OpenAI	o3-mini	1.10	4.40
Google	Gemini 1.5 Pro	1.25	5.00
Google	Gemini 1.5 Flash	0.075	0.30
Mistral	Mistral Large	2.00	6.00
Mistral	Mistral Small	0.20	0.60

Free vs Pro vs Team — What Each Plan Unlocks

Every formula in this guide is free to read and understand. The table below shows what changes when you upgrade to the interactive ledger.

Feature	Free	Pro ($19)	Team ($49)
Read this guide	✓	✓	✓
Copy the plain-text summary	✓	✓	✓
Feature rows in ledger	3	Unlimited	Unlimited
Scenarios	1	3	6
Calls/action & batch columns	✓	✓	✓
Model swap comparison	✓	✓	✓
Sensitivity grid	✓	✓	✓
Optimization suggestions engine	Preview	✓	✓
Copy CSV export	✓	✓	✓
JSON import / export	✓	✓	✓
Print / PDF export	✓	✓	✓
Copy full plan (plain text with formulas)	—	✓	✓
Shareable plan links	—	—	✓
Priority email support	—	—	✓

Every feature in the guide is visible to every tier. Pro and Team simply unlock higher caps and additional export formats inside the interactive calculator.

Every Formula Used to Calculate Results

The calculator uses twelve formulas. All rates are entered in USD per 1,000,000 tokens. All costs come out in USD, then convert to your chosen currency for display. Every formula below is shown in full — nothing is hidden.

1. Effective Input Rate (After Caching)

Where cacheFraction = cache% ÷ 100. Cached input tokens cost approximately 10% of the standard rate.

Example: at inRate = $3.00 and cache% = 80:

2. Cost Per LLM Call

3. Row Cost Per User Action

4. Row Monthly Cost

5. Total Monthly AI Cost

6. Cost Per Customer

7. Gross Margin

8. Break-Even Customers

9. Budget Projection, Burn Rate, and Days to Budget

10. Sensitivity Grid

For each combination of request multiplier rm ∈ {0.5, 1, 2, 4} and output-token multiplier om ∈ {0.5, 1, 2}:

Cells are shaded green, amber, or red based on position between the grid's minimum and maximum.

11. Model Swap Comparison

The interactive calculator applies these against four alternative models — Claude Sonnet, Claude Haiku, GPT-4.1 mini, and Gemini 1.5 Flash — using your exact row inputs (tokens, calls per action, cache %, and batch setting).

12. Optimization Suggestion Values

Each suggestion computes a hypothetical alternative and reports the saving:

Suggestions are sorted by saving descending; the top five are shown, covering: model swap, prompt caching, batch API, output-token reduction, and multi-call workflow consolidation.

Worked Example — Chat Assistant on Claude Sonnet

Visual — How Cost Flows Through the Calculator

What Is This Calculation Used For?

Per-feature AI cost estimation is used to answer six recurring questions during product planning.

Pricing decisions

If the AI cost per customer exceeds the monthly plan price, the product loses money on every user.

Feature feasibility

Some features are economically viable only on a cheaper model. The Model swap comparison tells you whether switching models turns an unprofitable feature into a profitable one.

Budget planning

Founders and finance teams set a monthly AI spend ceiling. The Budget guardrails panel projects month-end total and shows when the cap will be hit.

Engineering trade-offs

Prompt caching, batch mode, and output-length reduction are the three biggest levers. The Optimization suggestions panel quantifies each one.

Investor and stakeholder reporting

The Copy full plan and Export JSON buttons produce a portable summary of the assumptions and outputs.

Post-launch re-forecasting

When real usage data comes in, update the tokens and request counts to compare actual vs. plan.

Common Mistakes

These are the ten most frequent input errors. Each one is easy to make and produces a materially wrong result.

Mistake	What goes wrong	Fix
Entering monthly tokens instead of per-call tokens	Inflation by orders of magnitude	Tokens are per call. Monthly tokens go only as a count of actions.
Forgetting calls per action on agent workflows	Underestimates by 3–10×	If a user action triggers plan, search, and summarise steps, set callsPerAction = 3.
Assuming 100% cache hit rate	Over-optimistic by 10× on input cost	80% is aggressive. 40–60% is realistic.
Enabling batch mode for latency-sensitive features	50% saving, but responses arrive in hours	Only enable batch for async tasks.
Using rates from memory	10–30% errors vs. current pricing	Always copy the provider's published rates on the day you plan.
Ignoring other infra costs	Overstates gross margin	Include hosting, database, storage in Other infra cost.
Setting plan price equal to AI cost	Zero margin after other costs	Aim for AI cost ≤ 20% of plan price.
Running a single scenario only	Misses growth and stress cases	Use the Volume dropdown to test 0.5×, 1×, 2×, and 4× traffic.
Confusing user actions with API calls	Undercount by the workflow depth	User actions are clicks. API calls are what your code sends.
Assuming input and output rates are the same	Output typically costs 3–5× more	Enter each rate separately.

Accuracy Note — Read Before Relying on Results

Where accuracy is high

The arithmetic is deterministic. If your inputs match reality, the outputs will match your invoice to within rounding.

Where accuracy drops

• Token estimates: real prompts vary per user. Use median values.
• Cache hit rate: depends on prompt repetition. Verify with a small sample before scaling.
• Model pricing: providers change rates frequently. Re-verify before each planning cycle.
• Batch discount: assumed at 50% across providers.
• Cached token discount: assumed at 90% off.
• Currency rates: fixed defaults, editable in the UI. Not live FX.

Validation built into the calculator

• Rates above $500/1M tokens trigger a warning.
• Token counts above 500,000 per call trigger a warning.
• Request volumes above 10M/month trigger a warning.
• Rows with zero requests but non-zero tokens trigger a warning.
• Cost per customer exceeding plan price triggers a red error.
• Negative or sub-20% gross margin triggers a warning or error.

Backup and recovery

The calculator saves automatically to your browser's localStorage. Clearing site data (or using a private browsing window that closes) will wipe the plan. Use Export JSON after each major session for a portable backup you can Import JSON later.

Real-World Usage

Indie SaaS founder preparing a launch

Before launching a chat assistant, estimate cost at 50, 200, and 1,000 users. Check that break-even customers stays below your growth plan.

Agency scoping a client AI feature

Model each deliverable as a row, apply a 2× volume multiplier for peak months, and export the JSON to attach to a proposal.

Small product team evaluating a model migration

Use the Model swap comparison to see the delta between your current model and three alternatives.

Internal platform team planning budget

Model each internal AI service as a row, enter the company-wide budget cap, and use Days until budget hit to set an enforcement trigger.

Consultant preparing an AI cost audit

Recreate the client's current setup, then flip the cache %, batch, and model settings one at a time.

Frequently Asked Questions

Plans — For Readers Who Want the Interactive Calculator

Next Steps

Load your three most expensive features into the interactive ledger, run the sensitivity table at 2× volume, and open the optimization suggestions panel. Those three actions typically surface 20–40% of recoverable spend in the first planning session.

Everything in this guide is free to read and copy. If the interactive calculator saves you a planning cycle, the plans above cover the cost.

Last reviewed for clarity and accuracy. Rates and provider capabilities change frequently — verify before relying on the output for decisions.