Jev pricing: what a real workload actually costs
Jev charges $0.042 per million input tokens and nothing for output. That is the easy part. The calculator below also works out the three things that decide your real bill: whether you batch your questions, whether the rate limits let you finish in time, and whether your state fits in the context budget.
Fine-tune the workload
Tickets, rows, requests or documents you evaluate.
The content you send. Roughly 4 characters per token.
How many typed questions you ask about the same state.
Instructions plus option descriptions for one question.
Only affects the LLM columns. Jev never bills output.
Everything is computed in your browser. Nothing is sent anywhere, and the URL updates as you type — so the Share button sends someone your exact scenario, not just the homepage.
Fan-out saving: one call, many questions
Unique to this calculatorJev reads the state once and answers every question against it in parallel. Ask your 4 questions in one request and you pay for the item’s tokens once. Ask them in separate requests and you pay for them 4 times.
Throughput
Wall clock to push one day of volume at the published limits (250K tokens/s, 1,200 req/min).
- If token-bound
- 4.9 min
- If request-bound
- 1.4 h
Context budget
64,000 tokens per request in total, and 32,000 for the state plus the single longest question.
Fits. At this state size you could pack up to 1,058 questions into one call.
Same workload on an LLM
Both sides answer all 4 questions for an item in one call, so this is a like-for-like comparison. The LLM additionally pays for 50 output tokens per item. Every rate is editable.
- Jev 1.13 TypeSafe AI$93.24Output free · baseline 1x
- GPT-5.6 Luna OpenAI$624.006.7x vs Jev
- Claude Haiku 4.5 Anthropic$2,97032x vs Jev
- Claude Sonnet 5 Anthropic$5,94064x vs Jev
- GPT-5.6 Terra OpenAI$6,24067x vs Jev
- GPT-5.6 Sol OpenAI$11,880127x vs Jev
- Claude Opus 5 Anthropic$14,850159x vs Jev
- Claude Fable 5.1 Anthropic$29,700319x vs Jev
- GPT-6 Astra OpenAI$29,700319x vs Jev
Jev’s rate and limits come from docs.typesafe.ai/models, checked 2026-09-21. LLM rates are list prices at standard tier and exclude prompt caching, batch discounts and negotiated terms. This compares price only, never accuracy — whether Jev is accurate enough for your task is something to measure on your own labelled data.
Jev pricing and limits at a glance
| Model | Jev 1.13 (jev-1.13.0) |
|---|---|
| Input price | $0.042 per 1M tokens ($42 per 1B) |
| Output price | Free — output tokens are not billed |
| Per-request fee | None |
| Rate limits | 250,000 tokens/sec and 1,200 requests/min |
| Context | 64k tokens per request; 32k for state + longest question |
| Input types | Text only (string, JSON object, or array of text) |
| Aliases | jev-latest and jev-preview both resolve to jev-1.13.0 |
Source: docs.typesafe.ai/models. TypeSafe warns that rate limits are adjusting dynamically during early access.
Three things that move your Jev bill more than the rate does
1. How you batch
Jev reads the state once and answers every question against it in parallel. Four questions in one call pays for the item once. Four separate calls pays for it four times, for the same answers. See the fan-out maths.
2. Which limit binds
250,000 tokens/sec and 1,200 req/min are different ceilings. Small states hit the request limit long before the token limit. Work out which one you hit.
3. Output tokens
An LLM that reasons before answering can spend more on output than input. Jev has no output bill at all, which is where most of the gap comes from. Compare against a specific model.
Where you can call Jev
64k context per request. Early-access waitlist.
Same $0.042 / 1M input. Lists a 32K context window.
No separate TypeSafe signup if you already build on Vercel.
Pass-through /typesafe endpoint; clients use a LiteLLM virtual key.
Jev versus a specific model
Luna is the cheapest frontier-family LLM most teams reach for when classifying. Here is what the same workload costs on each, with output tokens counted.
Haiku is the usual baseline for high-volume classification and routing. LiteLLM measured Jev classifying 5.43x faster at 96% lower cost on their router corpus.
Sonnet is the mid-tier default in a lot of pipelines. On pure classification volume the gap is mostly output tokens, which Jev does not bill at all.
The frontier end of the comparison. This is where TypeSafe's headline savings ratios come from, so treat them as an upper bound rather than a forecast.
Jev pricing questions
How much does Jev cost?
Jev 1.13 costs $0.042 per million input tokens, which is $42 per billion. Output tokens are free and there is no per-request fee, so your bill is purely a function of how much state and how many question definitions you send.
Are Jev's output tokens really free?
Yes. TypeSafe bills input tokens only. Responses still report usage.output_tokens so you can see them, but they are not charged. This is why Jev's advantage grows against any LLM that reasons out loud before answering.
Is it cheaper to ask several questions in one Jev call?
Almost always. Jev ingests the state once and evaluates every question against it in parallel, so N questions in one call pays for the state once instead of N times. With a 500-token item and four questions, batching cuts the bill by roughly 70 percent.
What are Jev's rate limits?
250,000 tokens per second and 1,200 requests per minute for jev-1.13.0. Exceeding either returns HTTP 429. TypeSafe warns these limits are adjusting dynamically during early access, and higher limits are available on custom and enterprise plans.
What is Jev's context limit?
64,000 tokens per request for the state plus all questions combined, and 32,000 tokens for the state plus the single longest question. Requests over either budget will not fit and need to be split.
Is Jev cheaper than GPT or Claude?
For classification, scoring and routing, yes, usually by one to two orders of magnitude, because Jev charges only for input while an LLM charges for input and output. The exact ratio depends on your state size and how many output tokens the LLM needs; this page's calculator works it out for your numbers.
Where can I call Jev besides the TypeSafe API?
Jev is served through OpenRouter as typesafe/jev-1.13, through Vercel AI Gateway, and through LiteLLM's /typesafe pass-through endpoint. Prices match TypeSafe's direct rate, but limits can differ — OpenRouter, for example, lists a 32K context window against the 64K TypeSafe documents.
Is this the Japanese encephalitis virus?
No. This site is about Jev, the System One decision model from TypeSafe AI. JEV is also an abbreviation for Japanese encephalitis virus, which is a completely unrelated medical topic.