The moment you send a message to an AI, the meter starts running. Every fragment of text — yours and the AI's — clicks the counter. This guide teaches you to read the meter, choose the right vehicle, and send smarter directions.
Welcome to Tokens for Partners.
You'd expect the taxi meter to tick per word. It doesn't. It ticks per sub-word fragment — a unit called a token, decided by the model's internal vocabulary. Short common words cost one click. Long or rare words split into multiple clicks.
This matters because verbose, wordy prompts rack up the meter faster — not just in word count, but because complex vocabulary splits into more fragments. Technical jargon, long words, and multiple languages all cost more per character.
Every cab ride has two costs: the briefing before you set off, and the actual miles driven. AI works the same way. What you send to the model costs one rate. What it generates back costs another — and it's significantly higher.
Type any text — a client email, a prompt, a paragraph from a report. See exactly how many meter clicks it generates, and what it costs across different vehicle classes.
The single most powerful lever in AI cost strategy is model selection. Just as you wouldn't charter a helicopter to fetch a file from the next office, you shouldn't use the most expensive model for every task. Match the vehicle to the errand.
Claude Haiku 5.5 / GPT-6 Luna / Gemini 3.5 Flash-Lite. Nimble, affordable, perfect for short predictable routes. Not glamorous — but brutally efficient at scale.
Claude Sonnet 5.5 / GPT-6.1 Sol / Gemini 3.8 Flash. The everyday workhorse. Handles complex reasoning, nuanced writing, multi-step analysis — without the premium price of the top tier.
Claude Fable 5.1 & Opus 5.5 / GPT-6 Astra / Gemini 4 Argon. Extraordinary capability, extraordinary cost. When the problem demands the deepest possible reasoning — and budget is secondary.
When a client deploys AI at scale, they're not taking one cab ride — they're running a fleet of thousands of rides a day. Use these sliders to model a real enterprise AI deployment and see what lands on the annual invoice.
How you phrase your directions to the driver changes the meter reading dramatically — even for the same errand. Three people can ask for the same analysis and generate three very different token bills.
Every AI model has a context window — the total tokens it can hold in working memory at once. Think of it as the driver's mental dashboard. Overfill it, and they start forgetting earlier parts of the journey. Manage it strategically.
Imagine your cab driver's working memory is a 1,000,000-click dashboard (roughly what today's leading models hold). As the ride progresses, the dashboard fills up: your initial directions, the map you shared, everything said during the journey, space reserved for what they'll say next. When it fills up, they start forgetting what you said at the beginning. At enterprise scale, this means your AI starts giving incomplete answers.