Read the explanation
The default system prompt is eight hundred tokens, user message four hundred and two memory steps at two hundred fifty each. Base context is seventeen hundred tokens. Three tool calls of nine hundred tokens add twenty seven hundred per iteration and a completion adds five hundred. At one pixel per ten tokens base measures one hundred seventy, tool batch two hundred seventy and completion fifty. These are assigned counts, not tokenizer measurements. The loop maintains context by adding the completion and tool batch after each iteration, then also adds the new tool batch when counting input after the first. Across four iterations the counted input values are seventeen hundred, seventy six hundred, ten thousand eight hundred and fourteen thousand, totaling thirty four thousand one hundred. Completions total two thousand, so combined tokens are thirty six thousand one hundred. At one pixel per one hundred tokens input measures three hundred forty one, output twenty and combined three hundred sixty one. The cost breakdown labels twenty six thousand input as prompt memory and history and eighty one hundred as tool responses. These categories implement this simulator formula; they are not guaranteed to match actual provider accounting, caching or context trimming. With input priced eighty cents per million and output four dollars, input costs zero point zero two seven two eight dollars and output zero point zero zero eight. Total per run is zero point zero three five two eight. At eight thousand pixels per dollar their bars measure two hundred eighteen point two four, sixty four and two hundred eighty two point two four. Two concurrent agents each at fifty daily runs mean one hundred daily runs, yielding three point five two eight dollars a day and one hundred five point eight four in a thirty day month. Selecting local zero prices sets token fees to zero but does not include hardware, electricity or maintenance. Native canvas drawing and input/reset behavior require actual parity checks.