What it costs¶
The price line, the budget check, and what the live suite cost.
At the end of this page you can read jev_usage, know which requests the package refuses before sending, and have the measured cost of everything on this site.
The price¶
Jev bills input tokens only: $0.042 per million, output free (docs.typesafe.ai/models, read 2026-09-29). The API reports token counts, not dollars, so the package computes cost from the price in strands_jev/questions.py and keeps a tally per client.
from strands import Agent
from strands_jev import ALL_TOOLS
agent = Agent(tools=ALL_TOOLS)
agent.tool.jev_ask(state="Refund me.", questions={"angry": "Is the writer angry?"})
print(agent.tool.jev_usage()["content"][1]["text"])
The summary line reads calls, questions, input tokens, mean latency and cost so far. reset=True zeroes the counters after reading, for measuring one step at a time.
The limits¶
Two documented limits (docs.typesafe.ai/api): 64,000 tokens per request, 32,000 for the state plus the longest single question. The client estimates tokens as characters divided by 4 and refuses a request that cannot fit, before anything is sent:
from strands_jev import BudgetExceeded, check_budget
from typesafe_sdk import Noul
try:
check_budget("x" * 200_000, {"q": Noul(instructions="Is it long?")})
except BudgetExceeded as exc:
print(exc)
Jev.ask runs the same check on every request, so a tool call that would exceed a limit comes back as status: "error" with this message and no bill.
The message names the estimated numbers and the fix: filter the state down, or split it into chunks and ask each chunk. The package adds two ceilings of its own, 128 questions per request and 500 items per many-item call, as guards on the bill rather than documented API limits; both live in questions.py.
What the live suite cost¶
Every number on Measured came from hatch run live on 2026-09-29: 210 requests, 135,192 input tokens, $0.005677 in total, about 0.6 cents. Mean latency 175 ms per request from a laptop, network included.
A request with a few hundred tokens of state and three questions is a few hundred tokens; at $0.042 per million that is a few hundredths of a cent. The cost that matters is the one you can see: jev_usage, and the input_tokens field on every result.