Jev Pricing Explained: $0.042 per Million Tokens

The price

	Jev 1.14 (jev-1.14.0) and Jev 1.13 (jev-1.13.0)
Input	$0.042 per million tokens ($42 per billion)
Output	Free
Context	64k tokens per request (32k for state plus any one question); 262k with the experimental long-context header, at the same price
Rate limits	Set per organization and shown in the developer console; TypeSafe stopped publishing fixed numbers on 14 October 2026

Source: TypeSafe models page, checked October 18, 2026. Both versions cost the same. Images, an experimental feature, are billed as input tokens at the same price; TypeSafe hasn’t yet said how many tokens an image uses. See the Jev 1.14 changelog entry.

What you’re billed for

You pay for the input tokens in each request: your state plus your questions (instructions and option descriptions). Each response reports usage.input_tokens, so you can track spend per call. Because Jev ingests the state once and answers every question against it in parallel, asking several questions about the same state in one call is usually cheaper than making several calls.

There is no charge for output. That matters more than it sounds: with an LLM, a classification that “thinks out loud” before answering can spend more on output than on input.

A worked example

Say you classify 100,000 support tickets a day, each about 500 tokens of text, and an LLM would need about 50 output tokens per ticket to answer. Over a 30-day month that’s 1.5 billion input tokens and, for an LLM, 150 million output tokens.

Model	Input cost	Output cost	Monthly total
Jev 1.13	1,500M × $0.042 = $63	$0	$63
GPT-5.6 Luna	1,500M × $0.20 = $300	150M × $1.20 = $180	$480
Claude Haiku 4.5	1,500M × $1 = $1,500	150M × $5 = $750	$2,250
Claude Sonnet 5	1,500M × $2 = $3,000	150M × $10 = $1,500	$4,500

At these volumes Jev is roughly 8× cheaper than the cheapest LLM in the table and 36× cheaper than Claude Haiku 4.5. The gap is smaller than TypeSafe’s headline numbers because it compares against smaller models. It grows quickly if your LLM uses reasoning tokens or you compare against a frontier model.

Price isn’t the whole story. These are only comparable if Jev is accurate enough for your task, and that’s something to measure on your own data.

Calculate your own costs

About TypeSafe’s comparisons

TypeSafe’s home page says Jev is “444.6× cheaper” and “193.6× faster” than LLMs, based on its own published workflow evaluations of System One tasks. TypeSafe itself says these are probably at the high end of what you’ll see in practice, so treat them as an upper bound, not a forecast. The home page also cites a 238× lower input price than Claude Fable 5.1, which matches list prices: $10 versus $0.042 per million input tokens.

Other ways to pay

Jev is also available through OpenRouter, which lists the same $0.042 per million input tokens with free output and TypeSafe as the provider, and through Vercel AI Gateway. Limits can differ from TypeSafe’s direct API: OpenRouter, for example, lists a 32K context window versus the 64K TypeSafe documents. Check the gateway’s own page if you route through it. For higher rate limits, TypeSafe has a rate limit increase request form; enterprise terms such as zero data retention go through its support request form.

Frequently asked questions

Are Jev's output tokens really free?

Yes. TypeSafe's models page says Jev is charged per input token and output tokens are free. Responses still report usage, but only input tokens are billed.

How is the 444.6x cheaper figure calculated?

It comes from TypeSafe's own published workflow evaluations comparing Jev with frontier LLMs on System One tasks. TypeSafe says the figure is likely at the high end of real-world savings. Your ratio depends on input size, output length, and which model you compare against.

What are Jev's rate limits?

TypeSafe no longer publishes fixed numbers. Since 14 October 2026 its models page says limits are set per organization, in tokens per second and requests per second, and your organization's current limits are shown in the developer console. Before that it listed 100,000 tokens per second and 40 requests per second. Higher limits are available on custom and enterprise plans.