← AI Coding Plan Cost Calculator
Checking pricing pages yourself gets you the sticker prices: a flat monthly plan price on one page, and per-token API rates on another. That is a fine first step, and it is enough if all you want is the headline numbers. It does not answer the question that actually decides whether the plan is worth it: given your usage, which option is cheaper this month, and how far would that usage need to move before the answer flips.
This calculator takes the same numbers a pricing-page lookup gives you — the plan price, and the four separate API rates for input, output, cache write, and cache read tokens — plus your own token counts (or a light/typical/heavy starting point if you don't know them), and does the comparison arithmetic and the break-even math for you. It doesn't replace reading the pricing pages; the rates it starts you from come straight from them. It replaces the multi-step arithmetic afterward.
| Question | Where to answer it |
|---|---|
| I just want to see the sticker prices — the plan cost, or the per-token API rates. | The vendor's own pricing page — that quick lookup is exactly its job. |
| I want to know whether my plan is actually cheaper than the API for how I use it, and by how much. | This calculator — runs the four-rate, four-token comparison against your own numbers instead of a blended guess. |
| I want to know how much my usage would need to change before the cheaper option flips. | This calculator — reports the break-even usage level, which a pricing page has no way to tell you. |
| I don't know my own token counts and don't want to dig them up first. | This calculator's light/typical/heavy presets — a starting point you edit, not a blank form. |
In practice most people do both: check the vendor's pricing page for the current rates, then bring those numbers here instead of multiplying four rates by four token counts by hand and hoping the cache-read weighting didn't get lost along the way.
You can, and the sticker prices alone take thirty seconds to find. What they don't give you is the answer to the actual question: given how much you personally use the tool, which one is cheaper, and by how much would your usage need to change for that to flip. That requires four separate API rates (input, output, cache write, cache read), your own token counts across the same four categories, and the arithmetic to compare a flat monthly number against a usage-weighted one. Most people stop at the sticker prices because the rest is tedious to do by hand, not because it doesn't matter.
It runs the arithmetic you'd otherwise do by hand: multiplies your input, output, cache write, and cache read tokens by the matching API rate, compares the total to your plan's flat price, and reports the break-even usage level -- how far your usage would need to move before the cheaper option changes. It also starts you from a light/typical/heavy usage preset if you don't already know your own token counts, which most people checking pricing pages by hand don't.
The rate is on the page; the mistake is in the mental math, not the lookup. Publicly reported usage breakdowns for heavy agentic coding put cache reads above 90% of total tokens processed, because these tools resend most of the file and conversation context on nearly every turn. A back-of-envelope estimate that blends input and output at a single rate -- the shortcut most people take when doing this by hand -- significantly overstates the API-equivalent cost for exactly the heavy users asking the question. Keeping four rates and four token counts straight through a multi-step comparison is exactly the kind of arithmetic a calculator does more reliably than a person doing it in their head.
No. All arithmetic happens in your own browser, the same as if you were doing it on paper -- there is no account, no server call, and no analytics vendor. The only difference is the calculator does not mis-add the columns.