TokenCost
LLM token cost calculator
Paste tokens in and out. See estimated USD. Filter, copy a link, share the same comparison.
Compare API cost
Enter tokens per request. Optional daily volume projects a 30-day month. Sorted cheapest first for this mix.
Amazon Nova LiteCheapest
Amazon
$0.000240
this run
$0.7200/ mo
GLM-5.3-Flash
Z.ai
$0.000550
this run
$1.65/ mo
Llama 4 Scout
Meta (Bedrock)
$0.000670
this run
$2.01/ mo
GPT-5.6 Luna
OpenAI
$0.001000
this run
$3.00/ mo
DeepSeek V4 Flash
DeepSeek
$0.001540
this run
$4.62/ mo
Mistral Large 3
Mistral
$0.001750
this run
$5.25/ mo
Gemini 3.8 Flash
Google
$0.003375
this run
$10.13/ mo
Claude Haiku 4.5
Anthropic
$0.004500
this run
$13.50/ mo
Grok 4.6
xAI
$0.007000
this run
$21.00/ mo
Claude Sonnet 5
Anthropic
$0.009000
this run
$27.00/ mo
Gemini 3.1 Pro
Google
$0.0100
this run
$30.00/ mo
GPT-5.6 Terra
OpenAI
$0.0100
this run
$30.00/ mo
Kimi K3
Moonshot
$0.0135
this run
$40.50/ mo
GPT-5.6 Sol
OpenAI
$0.0180
this run
$54.00/ mo
Claude Opus 5
Anthropic
$0.0225
this run
$67.50/ mo
This run is one request at the token counts above. Per day multiplies by requests per day. Monthly multiplies that daily total by 30. Cheapest for this mix: Amazon Nova Lite.
Prices last updated September 8, 2026 (). Rates change without notice. Figures are estimates from public list prices for standard, uncached, real-time text tokens — not an invoice. Cached tokens, batch discounts, thinking tokens you did not enter, and image/audio/video usage are not modeled in v1. GPT-6 Astra (and GPT-5.6) prompts with more than 272K input tokens reprice the entire request at the long-context column — Astra becomes $20 / $75. Grok 4.6 prompts at or above 200k bill the whole request at $4 / $12. Confirm on the provider’s official pricing page before you budget.
Why this calculator
Built for English Google search in the US: compare an OpenAI API cost calculator, a Claude API pricing calculator, and a Gemini API cost calculator in one table.
Large language model APIs bill on tokens, not on “messages” or words. Input tokens are everything the model reads — system prompt, retrieved documents, chat history, tool results. Output tokens are what it writes. Providers publish a pair of USD prices per million tokens. This site multiplies your counts by those list rates. That is the entire v1 model: no login, no live scrape, one typed config file.
The packed snapshot holds about thirty-five public SKUs. The calculator opens on a shortlist; use Show all models plus the All / OpenAI / Anthropic / Google / Mistral / Amazon / xAI / China filters. Amazon includes Nova plus Llama 4 on Bedrock. China is Chinese labs (Z.ai, Tencent TokenHub, Xiaomi MiMo, Moonshot, MiniMax, DeepSeek, Alibaba) — not a mix of Western APIs.
Three things this calculator will not catch: prompt-cache hits, batch/flex discounts, and non-text meters (images, audio, video, search grounding). Those can move a bill by 2× or more. The cheapest badge is also not a quality badge — run evals. Use the pricing table when you need a crawler-friendly dump of the snapshot.
Frequently asked questions
How does this LLM token cost calculator work?
Enter input tokens, output tokens, and optional requests per day. TokenCost multiplies those counts by each model’s public list price per million tokens. You get a per-request estimate and, if monthly projection is on, a 30-day total. The table sorts cheapest first.
What is a token in an LLM API?
A token is a chunk of text the model reads or writes — roughly 4 characters or about 0.75 words in English, though code and other languages tokenize differently. Providers bill input tokens (your prompt plus context) and output tokens (the completion) at separate rates. Output is almost always more expensive.
Are these official, live prices?
No. Prices are a snapshot of public list rates stored in one config file and labeled with a last-updated date. Providers change SKUs, cache discounts, and batch rates without notice. Treat every figure as an estimate and confirm on the provider’s pricing page before you commit budget.
Does the calculator include cached tokens, batch discounts, or images?
Not in v1. Estimates use standard, uncached, real-time text rates. Prompt caching, Batch APIs (often ~50% off), thinking/reasoning tokens billed as output, image/audio/video tokens, and tool or search surcharges are called out in copy but not modeled.
How do I share a comparison?
Click Copy link. The URL encodes tokens, daily volume, toggles, and the selected models as query parameters so a teammate opens the same table.