The Million-Token Window, in Real Units
One million tokens is roughly 700,000 English words — around eight novels, 30,000 lines of code with documentation, or an hour of video transcribed. Almost nothing you paste into a browser tool approaches it, which is the point of the fit meter: it shows how much genuine headroom long-context workflows have. The practical constraints are cost and latency, since every token in the window is processed — and billed — on each request.
Why This Is an Estimate — and How to Get Exact
Google hasn't released Gemini's tokenizer vocabulary, so no third-party counter can reproduce its exact splits. Google's documentation offers the 100-tokens-per-60-to-80-words rule of thumb, and this tool's BPE-based estimate is finer-grained than that. When you need the real figure, the Gemini API's countTokens method returns it exactly, costs nothing, and doesn't invoke the model — worth wiring into any production pipeline.
Multimodal Tokens Count Too
Gemini prompts often include more than text, and everything is metered in tokens: Google documents images at a fixed 258 tokens each (tiled for large images), video at roughly 263 tokens per second, and audio at 32 tokens per second. If your prompt mixes media, the text estimate here is only part of the bill — add media at those documented rates.