Jev API Pricing: Cost per Token on All 4 Channels
Last checked · Independent guide, not affiliated with TypeSafe AI
Jev costs $0.042 per million input tokens ($42 per billion) and output tokens are free, on both TypeSafe's API and OpenRouter. A typical request with a short message and a few questions used 300 to 450 input tokens in our tests, which is about $0.00002 per call.
Jev’s pricing is simpler than an LLM’s: you pay for input tokens only. Everything Jev sends back, whether a choice, a score or a probability, is free.
Price by channel (September 19, 2026)
Section titled “Price by channel (September 19, 2026)”| Channel | Input | Output | Free option | Notes |
|---|---|---|---|---|
| TypeSafe API | $0.042 / 1M tokens | Free | Not published | $42 per billion tokens; enterprise plans for higher limits |
OpenRouter (typesafe/jev-1.13) |
$0.042 / 1M tokens | $0 | No | Same rate as TypeSafe; needs prepaid credits |
Vercel AI Gateway (typesafe-ai/jev) |
Listed as free | Listed as free | $5/month free tier credit | Promotional pricing ends Sep 25, 2026; card required |
Cloudflare Workers AI (typesafe/jev) |
Shown in dashboard only | — | Not stated for Jev | We have not been able to confirm a price |
TypeSafe’s launch post says it expects prices to go down rather than up, and admits it cannot prove its pricing is not subsidized. Treat current prices as early-access prices.
What real requests cost
Section titled “What real requests cost”These are input token counts from our own calls to the official API on September 19, 2026:
| Request | Input tokens | Cost at $0.042/M |
|---|---|---|
| One short message, one yes/no question | 287 | $0.000012 |
| One message, three questions (Noul, Choice, Score) | 451 | $0.000019 |
| One message, 20 yes/no questions | 559 | $0.000023 |
| A 26,867-token log file, one question | 26,867 | $0.0011 |
| The largest state we could send (32,204 tokens) | 32,204 | $0.0014 |
Two things stand out:
- Extra questions are almost free. Going from 1 to 20 questions on the same message added about 256 input tokens, roughly 13 tokens per question, and did not add latency. Asking everything in one call is much cheaper than one call per question, because the state is only paid for once.
- The state is the cost. A long document dominates the bill. Trimming the state to the part the question needs saves money and, according to TypeSafe’s own notes on Jev 1.13, improves accuracy.
For a sense of scale, a million requests like the three-question example above would cost about $19 in input tokens. Our earlier batch of 131 test questions on Vercel cost $0.005 in total, and 145 questions on the official API cost $0.0054.
How Jev compares with LLM pricing
Section titled “How Jev compares with LLM pricing”TypeSafe’s launch post puts LLM input prices at roughly $0.20 to $10 per million tokens, with output typically about five times the input price. Jev charges $0.042 for input and nothing for output. Two caveats keep this comparison honest:
- It only applies to jobs Jev can do: classification, routing, scoring and yes/no checks. If you need generated text, Jev cannot replace the LLM.
- The large multiples TypeSafe quotes (up to 444.6x cheaper) come from its own workflow evaluations, which it says are “on the higher end” of real-world gains. See Jev benchmarks explained.
Is there caching or a batch discount?
Section titled “Is there caching or a batch discount?”TypeSafe’s documentation does not mention input caching or batch discounts as of September 19, 2026. The closest thing is packing many questions into one request, which TypeSafe’s docs recommend and which our numbers above support.
Estimate your own cost
Section titled “Estimate your own cost”Cost per request ≈ (state tokens + roughly 15 tokens per short yes/no question, more for Choice and Score questions with long option descriptions) × $0.042 ÷ 1,000,000.
TypeSafe’s docs describe the 32k budget as about 150,000 characters of English, which works out to between four and five characters per token. Text full of numbers and IDs tokenizes less efficiently: our synthetic log file averaged about three characters per token. The usage.input_tokens field in every response gives the exact figure, so log it for a few days before projecting monthly spend.
Related
Section titled “Related”- Is Jev free? covers the Vercel promotion and free tiers in detail.
- Jev rate limits matter as much as price for high-volume jobs.
- Jev channels compared lists model names and context sizes per platform.