Token and cost numbers: measured vs estimated

What “tokens” means in Jenzo

Tokens are everything the AI processed for a piece of work, added together:

• New input: what was sent to the model fresh.
• Output: what the model wrote.
• Cached re-reads: earlier context the model read again from cache.
• Cache writes: context saved to cache for later.

Cached re-reads are usually the biggest part, because AI coding tools re-read your project many times. The AI Spend page shows all four parts so you can see where the volume comes from.

Measured

Measured numbers are exact. They come from the usage your AI tool already saves on your computer, read by the Jenzo helper or the Jenzo plugin for Claude Code. Each model call is priced per million tokens at the model vendor’s public list API prices, with separate rates for input, output, cache reads and cache writes.

You get measured numbers for Claude Code, Codex, Claude Desktop, Gemini CLI, OpenCode and Hermes once the helper is installed. See Connect your AI tools.

Estimated

Estimated numbers were reported by the AI tool itself when it stopped its clock, not read from its own usage files. Jenzo labels them reported by the AI, not measured. This happens when:

• the tool does not save token counts on disk (for example Cursor), or
• the Jenzo helper is not installed on that computer.

Estimates are priced at the model’s rates and are deliberately on the low side, so real cost is likely somewhat higher. Unmeasured work is never shown as free.

How they combine

When Jenzo has exact numbers for the same person, ticket and tool, the exact numbers win and the reported ones are ignored. Nothing is counted twice. Reports show the split, for example $X measured + ~$Y estimated or 85% measured. If the share of measured work is low, the page offers Set up exact measurement.

Why cost may differ from your AI bill

Jenzo shows cost at list API prices. If you pay a flat subscription for an AI tool, your bill will not move with these numbers, but they still tell you how much AI work each ticket, person and model used, in a unit you can compare. Models Jenzo does not know yet show tokens without a cost.