// instruments · autoswitcher
autoswitcher
Give it a job — how many tokens in and out, how many times, and how hard the task is — and it prices the job across the current Claude models, then picks the cheapest one that clears the tier you asked for. The lesson of the tool: don't pay frontier prices for a task a smaller model handles.
// the job
// how hard
most real work — summarize, draft, answer, light code
// the cost, per model
| model | tier | per call | × 1 |
|---|---|---|---|
| Haiku 4.5 | 1 ✕ | $0.00450 | $0.00450 |
| Sonnet 5← pickintro $2/$10 through 2026-08-31 | 2 | $0.0135 | $0.0135 |
| Opus 4.8 | 3 | $0.0225 | $0.0225 |
| Fable 5 | 4 | $0.0450 | $0.0450 |
For a standard job at this size, Sonnet 5 is the cheapest model that clears the tier — $0.0135 for 1 call. Paying for Fable 5 instead would cost $0.0450.
// honest about the inputs
Prices are $/1M tokens as published by Anthropic, current as of 2026-07 — they change, so verify before trusting a number. The tier map (Haiku=simple, Sonnet=standard, Opus=hard, Fable=frontier) is a working rule of thumb, not a spec: pick the tier by what your task actually needs, and test a cheaper one before assuming you need a dearer one. This is a calculator, not advice.