Jev Daily

Everton Carneiro says 201M tokens of Jev cost him $8.04

A lifetime spend figure for one workload, posted without a per-call price or an accuracy number.

Everton Carneiro posted his running Jev bill, $8.04 for 201 million tokens. Jev is the model doing keyword judging in Sextant, his product. His conclusion from that line item is that not every call in a stack needs a frontier model.

Everton Carneiro
@everton_dev
X
You don't need a frontier model for everything.
Nov 10, 2026 · View on X

That is the whole post. It is a screenshot of spend and one argument, and the argument is about model selection rather than about Jev specifically being good at judging keywords.

What the number is

It is cumulative spend to date on one workload, not a price per call and not a rate card. Carneiro does not say over what period the 201 million tokens accumulated, how many judgements that works out to, or how the input and output split. Keyword judging is the kind of task where the prompt dwarfs the completion, so a token count on its own does not tell you how many decisions got made for the eight dollars.

What it does tell you is the shape of the bill. For a classification-style call running at volume inside a product, the model line can sit in single digit dollars while the engineering around it costs more than that in an afternoon. That is the case Carneiro is making, and he is making it from his own invoice rather than from a pricing page.

What the number is not

There is no accuracy figure here. Nothing in the post says the keyword judging is correct, or how often it is wrong, or what it was compared against. Cheap and wrong is a real outcome, and this post cannot rule it out because it never measures the output. A spend screenshot is evidence about cost and only about cost.

"You don't need a frontier model for everything" is also an argument that only lands if you already know the task is in reach of the smaller model. Carneiro has presumably satisfied himself on that for keyword judging in his own product. A reader cannot inherit that judgement for a different task from this post.

For people building on it

If you are sizing a Jev workload from this, size the cost, not the fit. The useful takeaway is that a high volume judging call can be a rounding error on a monthly bill, which changes where you spend your attention. Whether Jev's judgements hold up on your keywords, your categories and your edge cases is a separate experiment, and this one does not run it.

Know what Jev is good at before you bet on it

One email a day. Under three minutes. What builders found out about Jev yesterday.

Get the next one by email

Jev, read daily so you do not have to. The builds, the benchmarks, the criteria that worked and the cases where it lost, from the people shipping on TypeSafe AI's System One model.