How current is the pricing?

Prices are tracked manually and dated at the top of the page. Last updated May 18, 2026 with Claude 4.x, GPT-5, Gemini 2.5, Nova, DeepSeek V4. If a provider changes pricing and we haven't updated yet, the per-1M figures at the right of the table show what's loaded — verify on the provider's site for critical decisions.

What does the 'At N calls' column mean?

Project the per-call cost to your monthly volume. Drag the slider to 100K to see what 100K calls of this text would cost at each model. Useful for sizing a production feature before you build it.

What does 'Input only / Output only / 50-50 mix' do?

LLM pricing splits input tokens from output tokens, usually with output costing 3-5x more. If the text you pasted is a prompt you'd SEND, pick 'Input only.' If it's a completion you'd RECEIVE, pick 'Output only.' For a rough total-cost estimate when you'll do both, pick '50/50 mix.'

Why isn't [my model] in the list?

We track the major frontier models from each provider. If you're using a specific Bedrock/Vertex/Azure model that isn't listed, find the closest peer model (Nova Pro ≈ similar to Claude Sonnet pricing-wise) for an estimate. Open an issue if there's a model with measurable user demand we should add.

Is anything sent to a server?

No. 100% client-side. Your text never leaves the browser. No tokenizer API calls, no logging, no analytics on the text content.

Need AI Training/Help?CloudYeti.io/meet

MarkdownMe

Token Counter & Cost Estimator (Claude, GPT, Gemini, Nova)

Paste any text and see token counts + USD cost across every major LLM provider — OpenAI (GPT-5, GPT-4o, o1), Anthropic (Claude Opus 4.5, Sonnet 4.6, Haiku 4.5), Google (Gemini 2.5 Pro/Flash), Amazon (Nova Micro/Lite/Pro), DeepSeek, xAI Grok, Mistral. OpenAI counts are exact (gpt-tokenizer / o200k_base BPE). Other providers use character-based estimates because they don't publish browser-runnable tokenizers — drift is clearly labeled. Choose input-only, output-only, or 50/50 mix; project cost at 1 to 1M calls. All client-side, no signup.

OpenAI counts are exact (using gpt-tokenizer, o200k_base BPE). All other counts are estimates because Anthropic, Google, Amazon, xAI, and Mistral don’t publish browser-runnable tokenizers. Estimates use a per-model character divisor; expect ±10% drift on prose, ±20% on code. Prices updated 2026-05-18.

Model	Tokens	Per call	At 1 calls	$/1M in/out
Claude Opus 4.5 Anthropic	66≈	$0.000990	$0.000990	$15.00 / $75.00
Claude Sonnet 4.6 Anthropic	66≈	$0.000198	$0.000198	$3.00 / $15.00
Claude Haiku 4.5 Anthropic	66≈	$0.000066	$0.000066	$1.00 / $5.00
GPT-5 OpenAI	47	$0.000470	$0.000470	$10.00 / $30.00
GPT-4o mini OpenAI	47	$0.000007	$0.000007	$0.15 / $0.60
Gemini 2.5 Pro Google	58≈	$0.000073	$0.000073	$1.25 / $10.00
Amazon Nova Micro Amazon	58≈	$0.000002	$0.000002	$0.04 / $0.14

≈ next to a token count means the value is estimated (character-based) rather than exact. Estimates use a per-provider character divisor: Claude ~3.5, Gemini/Nova ~4.0, Grok ~3.7, Mistral ~3.8. Real-world drift is typically ±10% for English prose and ±20% for code or non-Latin scripts.

Frequently Asked Questions

Are the token counts exact or estimates?: OpenAI counts are exact — we use gpt-tokenizer (o200k_base BPE), the same tokenizer GPT-4o/4.1/5/o1 use. All other providers (Anthropic, Google, Amazon, xAI, Mistral, DeepSeek) are estimates because they don't publish browser-runnable tokenizers. Estimates use a per-provider character divisor (Claude ~3.5, Gemini/Nova ~4.0). Drift is typically ±10% for English prose, ±20% for code or non-Latin scripts. The UI labels every estimate with a ≈ symbol so you always know which is which.
How current is the pricing?: Prices are tracked manually and dated at the top of the page. Last updated May 18, 2026 with Claude 4.x, GPT-5, Gemini 2.5, Nova, DeepSeek V4. If a provider changes pricing and we haven't updated yet, the per-1M figures at the right of the table show what's loaded — verify on the provider's site for critical decisions.
What does the 'At N calls' column mean?: Project the per-call cost to your monthly volume. Drag the slider to 100K to see what 100K calls of this text would cost at each model. Useful for sizing a production feature before you build it.
What does 'Input only / Output only / 50-50 mix' do?: LLM pricing splits input tokens from output tokens, usually with output costing 3-5x more. If the text you pasted is a prompt you'd SEND, pick 'Input only.' If it's a completion you'd RECEIVE, pick 'Output only.' For a rough total-cost estimate when you'll do both, pick '50/50 mix.'
Why isn't [my model] in the list?: We track the major frontier models from each provider. If you're using a specific Bedrock/Vertex/Azure model that isn't listed, find the closest peer model (Nova Pro ≈ similar to Claude Sonnet pricing-wise) for an estimate. Open an issue if there's a model with measurable user demand we should add.
Is anything sent to a server?: No. 100% client-side. Your text never leaves the browser. No tokenizer API calls, no logging, no analytics on the text content.

Related Tools

Markdown Cheat Sheet

Interactive Markdown cheat sheet with live examples. Free Markdown syntax reference guide.

Regex Find & Replace

Find and replace text in Markdown using regex. Free online regex find and replace with live preview.

YAML Frontmatter Editor

Visual editor for YAML/TOML frontmatter. Free online frontmatter builder for Jekyll, Hugo, and Astro.

Transcript Cleaner

Clean up auto-generated transcripts, SRT subtitles, and raw captions. Remove timestamps, filler words, and speaker labels. Free online transcript cleaner.

About Token Counter & Cost Estimator (Claude, GPT, Gemini, Nova)

Free cross-provider AI token counter and cost calculator. Exact OpenAI counts (gpt-tokenizer) + character-based estimates for Claude, Gemini, Amazon Nova, Grok, DeepSeek, Mistral. Compare prices side-by-side, scale to N calls per month. This tool runs entirely in your browser, so your content is not sent to a server. MarkdownMe provides free, fast, and reliable Markdown tools for developers, writers, and anyone who works with Markdown.