The short version
TypeSafe lists Jev at $0.042 per million input tokens — $42 per billion — with output tokens treated as free because the model does not generate strings.
On TypeSafe's published workflow evals that works out to roughly $0.0004 per decision case, against cents per case for frontier LLMs on the same workflows.
Why this matters
Cost per call is only interesting when you compare it to how often you call. The reason a decision model changes product design is that it makes the per-row case affordable: scoring every review, every message, every draft, rather than sampling a few.
It also changes budgeting shape. With a token-billed model, output length is the cost driver and prompts are written defensively to keep answers short. With free output, you can add questions without adding cost — only the state you send is billed.
How to model it
A simple four-line estimate beats a spreadsheet you never update.
- Estimate average input tokens per call, including the state you paste in
- Multiply by calls per month, then by $0.042 per million tokens
- Add the cost of escalation for cases below your confidence threshold
- Compare against the human minutes the decision currently consumes
The honest caveats
Vendor pricing is not proof of sustainable pricing, and early-access economics are frequently subsidised to seed adoption. TypeSafe says publicly that it expects prices to go down rather than up, but you should model a future where input pricing rises.
The other caveat is state size. Since only input is billed, the cheapest optimisation is not a shorter prompt — it is sending less irrelevant context. Trimming state is the real cost lever.
Common mistakes
Teams comparing models usually compare the wrong numbers.
- Comparing price per token instead of cost per completed decision
- Ignoring escalation cost in a confidence-threshold design
- Sending an entire record when three fields carry the signal
- Assuming today's early-access prices are the long-run prices
How PixaSocial Ai helps
PixaSocial Ai's plan model puts credits and AI jobs on one transparent surface, which makes it easier to reason about unit economics than a stack of per-seat tools. See plan credits or the articles hub.