AI tooling

GPT vs Claude vs Gemini: what tokens actually cost

Model list prices change often. Estimate the cost of a real prompt in your browser, with no API key and no need to paste your text into a third-party calculator.

LLM Cost Estimator Token Counter
Unit economics

Why token counts matter

LLM APIs bill by tokens, not characters. The same “10K-token prompt” can cost very different amounts depending on the model tier, the separate input and output rates, and whether part of the context is cached. Each vendor also uses its own tokenizer, so one piece of text can produce somewhat different token counts per model.

Input ≠ output

Completion tokens are usually priced higher. Always model both sides of a request.

Volume multiplies

A few cents per call becomes a real monthly bill when a feature hits every user.

Keep prompts local

Count tokens and estimate cost in the browser before you paste production text into a vendor console.

Playbook

Quick comparison workflow

1

Count tokens

Paste system + user text into the Token Counter.

2

Estimate cost

Open the LLM Cost Estimator and pick models.

3

Set volume

Enter input and output counts (or paste text), then project monthly spend.

4

Scan PII

Run the Prompt PII Scanner, or test with synthetic samples instead of real data.

Sanity check

Example: 10K in, 2K out

The math is the same for every provider: (input tokens ÷ 1,000,000 × input price) + (output tokens ÷ 1,000,000 × output price). With illustrative rates of $3 per million input tokens and $15 per million output tokens, a 10,000-token prompt with a 2,000-token reply costs $0.03 + $0.03 = $0.06. At 100,000 calls a month, that is $6,000. Real prices change, so check each provider's page below. The estimator uses published list prices, which helps when you review a vendor quote or the margins of an AI feature. To see several models side by side, use Cost Compare.

OpenAI GPT

Check current input/output rates and cached-input discounts on the official pricing page.

openai.com/api/pricing

Anthropic Claude

Compare the Haiku, Sonnet, and Opus tiers. On long generations, output tokens usually make up most of the bill.

anthropic.com/pricing

Google Gemini

The choice between Flash and Pro, and any pricing tiers based on prompt length, changes the cost of a 10K-token request far more than character count does.

ai.google.dev/pricing

Repeatable

Chain it in Recipe Studio

This follows the same count, cost, then scan pattern as Ship AI prompts safely. If a prompt runs close to a model's limit, check it with the Context Estimator before you chain it.

token-counter llm-cost-estimator prompt-pii-scanner

Open Recipe Studio

Estimate costs before you ship an AI feature

No API key needed, and your prompt text stays in the tab. Compare GPT, Claude, and Gemini using the same sample.

Client-sideToken counts and math run in the browser
Verify list pricesAlways re-check vendor pages