What is prompt caching and how does it reduce API cost?

Prompt caching stores pre-computed KV attention states for static context such as system instructions or background documents. Cache read hits bill at up to a 90% discount (e.g. $0.30/1M on Claude 3.5 Sonnet vs $3.00/1M uncached), drastically cutting operating expenses for agentic workflows.

How do Batch API discounts work?

For non-urgent workloads like bulk dataset processing, offline evaluations, or background document indexing, OpenAI, Anthropic, and Google offer Batch APIs that provide a flat 50% discount on both input and output token pricing with a 24-hour delivery turnaround.

AI Token & API Cost Calculator - Estimate LLM Token & API Costs

Multi-Provider AI Token & API Cost Engine Rates Verified: Sep 2026

What do you want to calculate? Choose workflow to jump directly

Model & Currency Setup

Caching Supported Batch Supported

Select LLM / Provider

Display Currency

Quick pick:

Tokens per Request

100% Client-Side

Input Tokens / Request

Prompt, system instructions & message history.

Output Tokens / Request

Generated completion & internal reasoning tokens.

Paste Sample Text / Prompt

Presets:

Characters

Words

Est. Tokens

Density Ratio

—

Script Mix: Latin: Indic: CJK: Arabic:

Transfer to Cost Workbench:

* Note: Token counts are estimates calculated 100% locally in your browser using script-aware character density rules. No prompt text is uploaded.

Volume & Discounts

Requests / Day

Operating Days / Month

Cached Input Percentage (%)

Batch Processing API (50% Off) Asynchronous 24h batch queues for non-realtime workloads.

Estimated Monthly Cost $225.00 $0.0075 Per Request

Estimated API Spend Verified

Estimated Monthly Cost

$225.00

Uncached Input 50%
Cached Input 0%
Output 50%

Per Request $0.0075

Daily Spend $7.50

Annual Total $2,700.00

Monthly Token & Cost Breakdown

Total Monthly Input Tokens: 45,000,000
Cached Input Tokens: 0
Total Monthly Output Tokens: 15,000,000
Combined Monthly Token Volume: 60,000,000

Uncached Input Spend: $112.50
Cached Input Spend: $0.00
Output / Completion Spend: $150.00
Net Monthly API Cost: $225.00

Benchmark Comparison

Side-by-Side Model Pricing Benchmark

Evaluated with your active prompt & volume inputs

Model & Provider	Input / 1M	Output / 1M	Cached / 1M	Cost / Request	Monthly Spend

Budget Forecasting

12-Month API Cost Growth Projection

Monthly Growth: %

Annual Projected Total

$3,842.10

Average Monthly Cost

$320.18

Month 12 Run Rate

$585.60

Month	Requests	Input Tokens	Output Tokens	Estimated Cost

Reverse Planner

Budget & Request Capacity Planner

Target Monthly Budget

Target Monthly Requests

Max Monthly Requests within Budget: 13,333
Max Daily Request Rate: 444
Max Monthly Token Capacity: 20,000,000
Required Budget for Target Requests: $375.00/mo

SaaS Unit Economics

AI Cost Per Active User & Margins

Active Users

Reqs / User / Mo

Sub Price / Mo

AI Infrastructure Cost per User: $0.38/mo
Total Monthly AI Infrastructure Bill: $375.00
AI Cost as % of Subscription Revenue: 1.88%
Contribution Margin per User: $19.62

Scenario Modeling

Conservative vs Base vs High Usage

Conservative

$62.50

12,000

Base Case

$225.00

30,000

High Scale

$675.00

75,000

Optimization Analyzer

Prompt Reduction & Savings Impact

Baseline (Before)

Optimized (After)

Monthly Savings

$180.00

Cost Reduction

44.4%

How AI Token & API Pricing Works

Tokenization Normalization

All modern LLMs price in units per 1,000,000 tokens ($/1M). For standard English, 1,000 tokens equals approximately 750 words (~4,000 characters). Output tokens generate higher computational overhead during autoregressive decoding and are priced 3× to 5× higher than input tokens.

Prompt Caching Economics

Prompt caching stores pre-computed KV states for static system instructions or document context. Cache reads bill at up to a 90% discount (e.g. $0.30/1M on Claude 3.5 Sonnet vs $3.00/1M uncached). Caching is automatically carved out from input tokens so you are never billed twice.

Batch API Queues

For asynchronous jobs such as document summarization, dataset classification, or offline evaluations, OpenAI, Anthropic, and Google offer Batch APIs with a flat 50% discount on both input and output tokens with a 24-hour turnaround SLA.

Good to know

Questions, answered

Quick answers about how this tool works.

? How are tokens calculated from text?

This calculator runs a local, script-aware character heuristic entirely inside your browser. It distinguishes English prose (~3.9 chars/token), code & JSON symbols (~3.3 chars/token), Devanagari/Indic scripts (~2.0 chars/token), and CJK ideographs (~1.5 chars/token). No text is uploaded to any external server.

? What are reasoning tokens in OpenAI o1 and o3-mini?

Reasoning models (such as OpenAI o1, o3-mini, and DeepSeek-R1) generate internal chain-of-thought reasoning tokens before producing the visible completion. These reasoning tokens are billed at the full output rate and count against your maximum completion limits.

? How current are the pricing rates?

Pricing shown here is based on cited vendor pricing documentation (OpenAI, Anthropic, Google Gemini, and DeepSeek official pricing pages) and may change when providers update their rates. You can also define custom model rates at any time.

? Is my prompt data private?

Yes, 100%. All arithmetic, token estimates, and scenario projections execute locally in client-side JavaScript. Your prompt text, token numbers, and financial parameters are never stored, logged, or transmitted.

Official Provider Pricing Sources

Official provider pricing references: OpenAI API Pricing (openai.com/api/pricing), Anthropic Claude Pricing (anthropic.com/pricing), Google Gemini Developer API Pricing (ai.google.dev/pricing), and DeepSeek API Documentation (api-docs.deepseek.com).

AI Token & API Cost Calculator