Skip to content

Free browser tools that respect your files: everything runs locally, nothing is uploaded.

Token & Cost Calculator

Estimate tokens and cost per run and per month across every supported model. No key needed — everything computes in your browser.

Your prompt

≈ 1 tokens (estimate)

Providers

17 models priced from the provider registry.

Cost comparison

ModelTokens inPer runPer month
Mistral Small 3.2 — cheap, strong defaultOpenRoutercheapest for this prompt135$0.0002<$0.01
GPT-5 nano — cheapest OpenAIOpenRouter135$0.0002<$0.01
Gemini 2.5 Flash-Lite — cheapestGoogle Gemini135$0.0002<$0.01
GPT-4o miniOpenRouter135$0.0003<$0.01
Llama 3.3 70BOpenRouter135$0.0005<$0.01
DeepSeek V3 — very cheapOpenRouter135$0.0006<$0.01
GLM 4.6 (Z.ai)OpenRouter135$0.0011$1.0675
Gemini 2.5 FlashOpenRouter135$0.0013$1.2905
Gemini 2.5 Flash — free tier workhorseGoogle Gemini135$0.0013$1.2905
Claude Haiku 4.5 — best writing quality per dollarOpenRouter135$0.0026$2.635
Claude Haiku 4.5 — fast, affordableAnthropic (Claude)146$0.0026$2.646
Gemini 2.5 ProGoogle Gemini135$0.0052$5.1687
Claude Sonnet 4.5 — best qualityAnthropic (Claude)146$0.0079$7.938

Estimates only: input tokens use a chars-per-token heuristic per provider family; prices are the registry's sampled list prices ( OpenRouter, Anthropic (Claude), Google Gemini, Ollama (local)). Your invoice is the only exact number — but prompt caching routinely cuts input cost 50–90% on repeated context.

Quick Answer

Paste a prompt (or leave the sample), set expected output tokens and runs per month, and the calculator shows every supported model ranked by per-run cost: estimated input tokens, cost per run, cost per month at your volume, and the cheapest model highlighted. Token counts are chars-per-token estimates per provider family and prices are providers’ published list prices from this site’s registry — close estimates, not an invoice. Prompt caching notes are included since cached input routinely costs 50–90% less on repeated context.

How It Works

1

Paste your prompt (or leave the sample) — input tokens are estimated per provider family

2

Set expected output tokens per run and your runs per month

3

Compare every model side by side: per-run cost, per-month cost, and the cheapest for this prompt highlighted

4

Remember prompt caching: repeated long contexts routinely cost 50–90% less than the shown input price

Key Facts

  • Every model in the site’s provider registry priced side by side
  • Per-run and per-month projections at your actual volume
  • Cheapest-model highlight updates as your prompt changes
  • Token estimates per provider family (chars-per-token heuristic, labeled)
  • No API key, no network — the whole calculation runs in your browser

Frequently Asked Questions

How accurate are the token counts?

They are estimates from a chars-per-token heuristic per provider family (real tokenizers vary with content). Costs inherit that estimate plus the provider’s list price. For exact numbers, your provider’s usage dashboard is the source of truth — this tool is for comparing options, which the heuristic is accurate enough for.

Does it account for prompt caching?

The estimate uses standard input prices; the page notes that cached input (Gemini and OpenAI explicit caching, Anthropic prompt caching) routinely cuts input cost 50–90% on repeated context — so long-system-prompt workloads come in below the shown number.

Where do the prices come from?

The same registry that powers this site’s bring-your-own-key tools, sampled from provider price pages. Prices change; the registry is updated when providers are re-verified.