knaxly.dev / gauge

Gauge token studio

Count tokens and estimate cost for GPT-5, Claude, Gemini and more. See if your prompt fits the context window. Free, private, and it runs entirely in your browser.

Prompt text Agent mode

System prompt

Tool / function definitions

Message history

In agent mode all four fields roll up into one total and are budgeted against the model context window.

Per-token view

Exact for OpenAI encodings

Type something to see how it tokenizes.

Model

Tokens

Characters

Words

Tokens / word

Context window

input 0 reserved output 0

Reserved output tokens

Estimated cost

$0 total

input $0 · output (512 tok) $0

🔒

Private by default. Your text never leaves your tab. Counting runs locally in your browser, with no network calls. Open DevTools and watch the Network panel while you type to confirm.

Compare across models

Same text, sorted by total cost

Model	Tokens	Context	In $/1M	Out $/1M	Total cost

Pricing and context limits (editable config)

Prices last updated: . These change often, so treat them as approximate. Values live in the MODELS config object in this file and are easy to edit. Prices are USD per 1M tokens.

OpenAI counts are exact (computed with the real GPT tokenizer in your browser). Claude, Gemini, DeepSeek and Llama counts are estimates marked with ≈: no public browser-side tokenizer exists for those, so Gauge takes an OpenAI count and applies a per-model calibration multiplier. Expect roughly +/- 10 to 15% for English prose, and wider for dense code or non-Latin scripts.