Tools · Cost of the job
What does AI actually cost you?
Not per million tokens — per month, for the work you actually do. Pick the closest job below. The arithmetic is printed underneath so you can disagree with it.
Drafting and editing long pieces — reports, posts, proposals.
Near-frontier workhorse, ~6–12 months behind.
Cheapest that can do it
£0.05 /mo
DeepSeek V4 Pro · $0.07
Dearest at the same tier
£0.63 /mo
GPT-5.6 Terra — 11.5× more for the same job.
How this was worked out
20 pieces a month · ~3,000 words of context in, ~1,500 words out each · 1.33 tokens per word.
That comes to 0.08M tokens in and 0.04M out per month, with 0% of the input assumed to be repeated context served from cache.
Prices are the provider’s own published rates, read from their pricing pages. Converted at £0.79 per $1 as of 2026-07-28. Your real bill will differ — these are the assumptions above applied to today’s list prices, not a quote. If an assumption is wrong for you, the arithmetic is right there to redo.
What surprises people
Personal use is nearly free; agents are the bill. One person writing and researching all month runs up a cost measured in pennies. The same person pointing an agent at a codebase all day is two or three orders of magnitude higher — because an agent re-reads its context on every turn, and turns are cheap to spend.
The tier matters more than the provider. Dropping from frontier to a strong or mid model usually changes the bill far more than switching vendors at the same tier. Most people shop vendors and never ask whether the job needed the clever model.
Caching is the lever nobody pulls. If your input repeats — the same codebase, the same knowledge base, the same system prompt — the cached rate applies and it is a fraction of the standard one. The presets above assume a realistic cache share for each job; that assumption is doing a lot of work.
Cost per token is the wrong metric. What matters is cost per completed task. A cheaper model that needs three attempts is dearer than an expensive one that lands first time, and no pricing page will tell you that — only your own evaluation will.
Prices are each provider’s own published list rates, read from their pricing pages and kept as a dated series (latest reading 2026-07-20). This is list pricing, not your negotiated rate, and it excludes anything you pay for hosting, retries, or the humans checking the output. The investor-facing view of the same dataset — the spread between open and closed weights, and how it has moved — is at the Token Price Index.