Which AI coding cost figures should this table track next?

[xyzs996]

The table in this repo currently holds 270 figures pulled out of 26 write-ups — prices, percentages, multiples, token counts and durations. Every row carries the full sentence it came from and a link to the piece, because a number on its own is not checkable: $1.43 could be per million tokens, per month, or per seat.

• Browse it: figures.md
• Read it in code: figures.json / figures.csv — those two always resolve to the newest dated snapshot, so a row you cite today stays where it is.

What it does not cover yet

Being blunt about the gaps, because they are the interesting part:

• No primary-source column. Each row links to the write-up that published the figure, not to the vendor page the figure came from.
• Prices only as of their own date. published is the day the write-up went out, not the day the price was in force. Model pricing moves fast enough that some rows are already historical.
• No per-provider rollup. You cannot currently ask "everything this table knows about Claude Code" without filtering yourself.
• Agent-run durations are thin. 75 duration rows, but almost all of them are single runs rather than repeated measurements.

The question

If you were pulling this table into something, which figure is missing that you would actually use? Concretely — a metric, a provider, a unit. "Cost per merged PR", "tokens per completed task by model", "how long a full-repo review takes at each context size", that kind of thing.

Replies here get turned into rows. If a figure is already in the table and it is wrong or stale, open a correction issue instead — that path is faster.

The data is CC BY 4.0. No account needed for any of it.