T TokenCost LabCompare LLM API costs across OpenAI, Anthropic and Google.
Español Go ad-free

LLM API Cost Calculator

Estimate what your AI feature will cost per month on OpenAI, Anthropic and Google, including prompt caching and batch discounts.

🔒 Everything is calculated in your browser. Your numbers never leave your device.

Your usage

Estimate tokens from a sample prompt
≈ 0 tokens
Cheapest / month$38GPT-6 Luna · per month
Most expensive$3,777GPT-6 Astra
Price spread100×most vs least expensive
Tokens / month232.5M150,000 requests

Monthly cost by model

Prices verified 2026-11-08
GPT-6 Luna$38
Claude Haiku 5.5$38
Gemini 3.1 Flash-Lite$108
Gemini 3.5 Flash-Lite$166
Gemini 3.8 Flash$283
GPT-5.4 mini$323
GPT-6.1 Terra$374
GPT-6.1 Sol$748
Claude Sonnet 5.5$748
Gemini 4 Argon$748
Gemini 3.1 Pro (preview)$860
Claude Opus 5.5$1,496
GPT-5.5$2,151
Claude Fable 5.1$3,723
GPT-6 Astra$3,777

Price table (USD per 1M tokens)

ModelInputCachedOutput1K requestsMonth
GPT-6 Luna budget$0.1$0.01$0.5$0.25$38
Claude Haiku 5.5 budget
Prompts ≤100K tokens. Above: $0.50 / $2.50.
$0.1$0.01$0.5$0.25$38
Gemini 3.1 Flash-Lite budget$0.25$0.025$1.5$0.72$108
Gemini 3.5 Flash-Lite budget$0.3$0.03$2.5$1.11$166
Gemini 3.8 Flash budget
Rises to $1.50 / $7.50 on Jan 1, 2027.
$0.75$0.075$3.75$1.89$283
GPT-5.4 mini budget$0.75$0.075$4.5$2.15$323
GPT-6.1 Terra workhorse$1$0.05$5$2.49$374
GPT-6.1 Sol workhorse$2$0.1$10$4.99$748
Claude Sonnet 5.5 workhorse$2$0.1$10$4.99$748
Gemini 4 Argon workhorse
Rises to $4 / $20 on Dec 1, 2026.
$2$0.1$10$4.99$748
Gemini 3.1 Pro (preview) flagship$2$0.2$12$5.74$860
Claude Opus 5.5 flagship$4$0.2$20$9.98$1,496
GPT-5.5 flagship$5$0.5$30$14.34$2,151
Claude Fable 5.1 flagship$10$0.25$50$24.82$3,723
GPT-6 Astra flagship$10$1$50$25.18$3,777

List prices for standard requests; long-context surcharges, regional pricing and taxes are not included. Always confirm on the provider pricing page.

How it works

  1. Pick a preset that looks like your app, or type your own traffic.
  2. Set the average input and output tokens per request. Not sure? Paste a sample prompt in the estimator.
  3. Add the share of input that hits the prompt cache, and switch on Batch if results can wait up to 24 hours.
  4. Compare the monthly cost of every model, sorted from cheapest to most expensive.

Why input and output are priced differently

Providers charge several times more for output tokens than for input tokens, because generating text is more expensive than reading it. Apps that write long answers (agents, content generation) are dominated by output cost; apps that read long documents (RAG, summarization) are dominated by input cost.

How prompt caching changes the bill

When the beginning of your prompt repeats between requests (system prompt, tool definitions, a long document), providers can serve it from cache at up to 90–97% off. A chatbot with a large fixed system prompt can often cut its input bill in half with caching alone.

Frequently asked questions

How accurate is this calculator?

It uses each provider's published list price per million tokens and the date they were last verified is shown above the chart. Your real invoice can differ because of long-context surcharges, regional pricing, taxes or negotiated discounts.

What is a token?

A token is a chunk of text of roughly 4 characters in English (about ¾ of a word). 1,000 tokens is about 750 English words. Spanish and code usually need a few more tokens per word.

Does batch processing work with caching?

Yes. All three providers offer 50% off for batch jobs, and cache-read discounts still apply on top. The calculator applies both.

Which model is the cheapest?

For most workloads the budget tiers (GPT-6 Luna, Claude Haiku 5.5, Gemini Flash-Lite) cost less than a tenth of the flagships. Use the table to compare quality tiers side by side.