How are LLM API tokens priced?

LLM providers charge per million tokens (1M tokens). Input (prompt) tokens are priced lower than output (completion) tokens. In English, 1,000 tokens equal approximately 750 words.

What is prompt caching discount in LLM APIs?

Prompt caching allows repeated prefix context (such as system instructions and RAG documents) to be cached in provider GPU RAM, reducing input token pricing by 50% to 90%.

LLM API Pricing & Token Cost Calculator

Estimate and compare API token expenses across OpenAI, Anthropic Claude, Google Gemini, and DeepSeek models.

Select AI Model

Input (Prompt) Tokens per Request

Output (Completion) Tokens per Request

Daily Requests / Invocations

⚡ Enable Prompt Caching (50% input discount) 📦 Batch API Discount (50% off total)

Estimated API Cost Projections Tier: Pay-as-you-go

Cost Per Single Request

$0.00875

Daily Projected Spend

$8.75

Monthly Projected Spend (30d)

$262.50

Annual Projected Spend

$3,193.75

Monthly Input Tokens

45,000,000 (45M)

Monthly Output Tokens

15,000,000 (15M)

💡 Cost-Saving Tip: Switching to mini or flash models (e.g. GPT-4o-mini, Gemini 2.0 Flash, DeepSeek V3) can reduce monthly API bills by up to 90%–96% while retaining strong classification and summarization accuracy.

LLM API Pricing Mechanics & Token Optimization

Large Language Model providers bill users based on token processing volume measured in units of one million tokens (1M). Understanding input vs. output cost structures, prompt caching, and batch processing allows cloud architects to slash AI operational costs.

Multi-Provider Model Comparison (Per 1 Million Tokens)

Model	Provider	Input / 1M	Output / 1M	Context Window
GPT-4o	OpenAI	$2.50	$10.00	128k
GPT-4o mini	OpenAI	$0.15	$0.60	128k
Claude 3.5 Sonnet	Anthropic	$3.00	$15.00	200k
Claude 3.5 Haiku	Anthropic	$0.80	$4.00	200k
Gemini 2.0 Flash	Google Cloud	$0.10	$0.40	1,000k
DeepSeek V3	DeepSeek	$0.14	$0.28	64k

Architectural Strategies to Lower LLM Costs

Model Routing & Cascading: Send simple classification, entity extraction, and formatting queries to ultra-cheap models ($0.10/M) and escalate to flagship reasoning models ($2.50+/M) only when needed.
Prompt Caching: Structure system instructions and retrieved context (RAG) at the beginning of the prompt to take advantage of provider prompt caching discounts.
Batch API Processing: For non-real-time evaluations, synthetic data generation, or nightly offline jobs, batch APIs provide an automatic 50% discount.