Token cost tools - simulator & visualiser

Two views on the same idea: every turn re-sends the whole conversation, so a session that drifts across unrelated tasks pays for work it finished with. Use the simulator to explore the shape of the curve, and the visualiser to price a concrete piece of work end to end.

Scenario

Model

Different tasks in one session 4

Turns per task 10

Tokens added per turn 3,000

Output tokens per turn 700

Baseline: system + tools + skills 15,000

Assumptions

Prompt caching on (cache read = 0.1× input, 5-min write = 1.25×) Charge output tokens (identical in both arms — uncheck to isolate the context effect)

Scale it up

Seats 200

Sessions like this per seat, per year 220

One long session

–

no /clear

Cleared per task

–

/clear on task switch

Saved

–

per session

Peak context

–

vs cleared

Context carried into each turn

One long session Cleared per task task boundary

Cumulative cost

One long session Cleared per task

Per-task breakdown

Task	Start context	End context	Cost - no clear	Cost - cleared	Delta

Scenario

Baseline model - used for every step in the left arm

model routing /clear subagents prompt caching

Baseline

–

Optimized

–

Saved

–

Peak context

–

History at the worst step

–

Baseline - context window at each step's last turn

Optimized - context window at each step's last turn

system + tools + skills accumulated conversation history tool output on this turn your prompt 200k context window

Bar heights reflect token volume, driven by /clear and subagents - model routing changes cost only (a different model's $/token), not the tokens shown here, so toggling it moves the KPIs above without moving these bars.

Step-by-step breakdown

Step	Turns	Baseline - one model, one session			Optimized - routed, cleared, delegated				Saved
Step	Turns	Model	Context at last turn	Cost	Model	Context at last turn	Subagent	Cost	Saved

Which lever did the work?

Applied cumulatively	Cost	This lever	Running total

Levers are applied in this order, so each row's credit depends on what came before - these are not independent contributions. Untick the toggles above to see the credit move.

Token cost tools - simulator & visualiser

Scenario

Assumptions

Scale it up

Context carried into each turn

Cumulative cost

Per-task breakdown

Baseline - context window at each step's last turn

Optimized - context window at each step's last turn

Step-by-step breakdown

Which lever did the work?

Reading this honestly