Jev Daily

Allie the Icon spells out how Jev charges state tokens

State is billed once per call no matter how many questions ride along, which changes when batching actually saves money.

Allie the Icon posted the input token math for Jev in a reply to @winstondu. The formula she gives is state plus the sum of all questions, and the part that matters is that state is charged once per call no matter how many questions run in parallel.

Allie the Icon
@allietheicon
X
you only get charged once for your state tokens no matter how many questions you run in parallel
Nov 14, 2026 · View on X

That has a concrete consequence. Two questions in one call do cost more than one question in that call, because you are paying for the extra question tokens. But asking those same two questions as two separate requests carrying the same state costs more overall, because the second request pays for the state again.

When splitting still wins

The interesting half is the case against batching everything. Allie says if you are more cost sensitive than speed sensitive, it is worth measuring how often each follow up question actually gets used. A follow up that is needed very rarely can be cheaper to run on demand, even with the duplicated state, than to attach to every call.

When speed matters more than cost, she says batching pretty much always wins, with one carve out. A single very large follow up question can slow down the whole call, so moving it into its own request can make sense depending on how often that answer is needed.

This is the pricing side of the pattern she described earlier, where you ask questions before you know you need them. The billing model is what makes that pattern cheap rather than wasteful, and it is also what bounds it.

The shared state point

Diogo Almeida made the same observation from the other direction, saying surprisingly few people realize state is shared and asking a lot of questions is very efficient. Neither post includes a measured bill or a token count, so this is the mechanism rather than a benchmark. Nobody here published a before and after number.

Diogo Almeida
@CompleteSkeptic
X
surprisingly few people realize state is shared and asking a lot of Qs is very efficient
Nov 14, 2026 · View on X

For people building on it, the thing to get right is that the decision is not batch or do not batch. It is a per question frequency question. Work out what fraction of calls actually consume each follow up answer, multiply that against the cost of the question tokens, and compare it to the cost of re-sending your state. If your state is small, splitting is cheap and the arithmetic barely moves. If your state is large, which is the usual reason people reach for this pattern at all, the duplicated state dominates and almost everything belongs in one call. Either way the number you need is your own state size, and neither post tells you what yours is.

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.