Jev pricing
Jev costs $0.042 per million input tokens, and output is free. There is no free tier. What that works out to on real work is the useful number: across 15 runs whose authors published both a volume and a bill, the median is $0.000065 per decision, or about 15,291 decisions per dollar.
Updated · by Jon Kraayenbrink
In short
- $0.042 per million input tokens. Output free. No free tier.
- Median published cost: $0.000065 a decision. Riley Brown’s 500 emails came to 3.5 cents.
- Questions are nearly free; the state is what you pay for. 61 questions about one post cost $0.0004.
- The comparison that matters is not per token. It is 100,000 posts for $0.67 against 214 posts for $0.98 on Claude Opus 5, in the same 20 seconds.
The published price
One price, one direction. TypeSafe bills the tokens you send and charges nothing for what comes back, because what comes back is a typed answer rather than text.
- Input
- $0.042 per million tokens
- Output
- Free
- Free tier
- None published
- Model
- Jev 1.13 and Jev 1.14, same price. jev-latest points to 1.13 and moves to 1.14 around 21 Oct 2026; jev-preview is 1.14
- Rate limits
- 100,000 tokens per second and 80 requests per second as of 8 Oct 2026. TypeSafe says the limits are adjusting and can change without notice; going over returns 429.
- Context
- 64k per request and 32k per question by default; 262k for both with the opt-in long-context header, same price
- Images
- Billed as input tokens at the same rate; no separate charge per image (preview)
- Latency
- 70 to 500 ms per call
Source: TypeSafe models and the launch post, which is the price every figure below was paid at. The Jev 1.14 release on 14 October 2026 left the price where it was; the long-context mode costs the same per token, and TypeSafe says it is slower and less accurate than the default. To try your own volume against it, use the LLM cost calculator.
What people actually paid
These are the runs where the author published a volume and a total, so the cost per decision is arithmetic rather than an estimate. They span 26 sheets of construction drawings to 123,805 API requests.
| The run | Volume | Total | Per decision | Per dollar |
|---|---|---|---|---|
| Every’s editorial vibe check | 1,709 passages judged | <$0.0100 | $0.000006 | 170,900 |
| 100,000 viral posts in 20.4 seconds | 100,000 posts scored | $0.6700 | $0.000007 | 149,254 |
| X timeline labeler | 1,000 posts labelled | $0.0300 | $0.000030 | 33,333 |
| Fraud detection with Jev and Kimi K3 | 100 emails checked | $0.0030 | $0.000030 | 33,333 |
| 3,282 posts, eight questions each | 3,282 posts analysed | $0.1282 | $0.000039 | 25,601 |
| 700 live ads in 40 seconds | 700 ads read | $0.0400 | $0.000057 | 17,500 |
| 1,891 ads in 19 seconds | 1,891 ads classified | $0.1200 | $0.000063 | 15,758 |
| 1,315 posts across eight dimensions | 1,315 posts sorted | $0.0860 | $0.000065 | 15,291 |
| Which exam questions come back | 377 questions rated | $0.0256 | $0.000068 | 14,727 |
| 500 emails for 3.5 cents | 500 emails triaged | $0.0350 | $0.000070 | 14,286 |
| 1kpapers | 1,018 papers classified | $0.0800 | $0.000079 | 12,725 |
| jev-evaluation | 123,805 requests | $12.6900 | $0.000102 | 9,756 |
| 724 competitor ads, broken down | 724 ads read | $0.0900 | $0.000124 | 8,044 |
| 700 leads scored in 40 seconds | 700 leads scored | $0.0900 | $0.000129 | 7,778 |
| A 26-sheet plan set in 2.9 seconds | 26 drawing sheets read | $0.0052 | $0.000200 | 5,000 |
Each row links to the build and to where its author published the figures. The full workings, with latency and throughput, are in the Jev Build Report.
Riley Brown
@rileybrown
Yeah Jev by @typesafeai is very cool. It classified 500 emails in seconds. And it costed 3.5 cents.
Movez
@0xMovez
I just built a Jev X Viral Post Analyser. 100,000 viral X posts. 20.4 seconds. $0.67. Claude Opus 5, same corpus, same clock, got through 214 posts and spent $0.98. per post that is ~680x cheaper the full Opus pass would have run $458. viral analysis is the perfect Jev job. • it is not writing, it is 14 yes/no calls per post: > does the hook open a loop, > is there a number in the first line, > is the proof real or claimed. classification, not prose. • what it found: 1,220 posts broke into the top 1%. baseline 1.22%. > superlative claim - 2.34% viral. 1.92x baseline > contrarian take - 1.59%. 1.31x > launch / tool drop - 1.46%. 1.19x and numbered lists, the thing everyone writes: 0.55%. below baseline. the most used hook is the least viral one. full stop. • what you are watching: left is the post under analysis, right is Jev answering 14 typed questions about it, each with a confidence score. the run stops at 20.4s because that is when Jev finished all 100k. pulled the corpus through a few X APIs, one parallel pass into Jev. should I drop it to public? Read my latest article on Jev Engineering below and turn your ideas into reality.
Hassan
@nutlope
I used Jev to classify 1,018 AI research papers. The result: $0.08 total cost and 256ms median end-to-end latency per paper. The pipeline was: 1. Summarize each paper with DeepSeek V4 Flash 2. Send the title + summary + 24 possible topics to Jev 3. Use Jev to classify each paper 4. Visualize everything on http://1kpapers.com The summaries cost $3.99 on @togethercompute. The classifications cost $0.08 on @typesafeai. So for just over $4 of inference, I ended up with a pretty useful way to explore the top AI research papers from the past year. I think this is where things are heading: different models for different parts of the workflow, instead of using one model for everything. I’m running evals on the Jev classifications before replacing the current ones, but the site is already live: http://1kpapers.com
Trinay Hari
@hari_trinay
Built a construction plan-set classifier with Jev. Proq turns civil and building plan sets into bills of materials using an LLM pipeline we built on GPT-4.1. Jev classified an entire 26-sheet plan set in 2.9 seconds for $0.0052. It matched GPT-4.1 and GPT-6 Astra on 100% of sheet-level classifications while running 17–21x cheaper and 5x faster than our production pipeline.
Worked cost example: the TypeSafe quickstart
The TypeSafe quickstart shows a support ticket with two questions: which team should handle it, and is it urgent. The call uses 392 input tokens.
- Input tokens
- 392
- Cost per call
- $0.0000165 (392 × $0.042 per million)
- Per 100k calls
- $1.65
- Per 1M calls
- $16.46
- Output cost
- Free
Real sizes vary. Putting several questions in one call is cheaper than one call per question: TypeSafe’s parallel-questions cookbook reports 12.2× cheaper and 10.0× faster for 13 questions in one call versus 13 separate calls.
Sources: TypeSafe quickstart, TypeSafe models, parallel-questions cookbook.
Your saving depends on how much of your bill is decisions
Jev can only replace the share of LLM spend that goes to classify, route, score and approve calls. Drafting, summarising and code stay on an LLM.
The ceiling on the saving is (decision share × LLM cost) − Jev cost. If 30% of your bill is decisions that Jev can handle, and your LLM costs $1,000, the saving ceiling is $300 minus what Jev costs for those decisions.
Gipp’s priced event desk demonstrates this: on his published prices, Jev costs $0.84 of a $29.04 day. Only about 4% of events reach a model that writes, so the 96% that Jev filters are where the saving comes from. More on the Jev router page.
Is Jev free?
No, and nothing on TypeSafe’s models page offers a free tier. The reason people ask is that the bills look like rounding errors: Mike Taylor ran 1,709 editorial judgments for under a cent, and Movez scored 100,000 posts on fourteen questions each for 67 cents. At the median rate, a dollar buys roughly 15,291 decisions.
If you want to try it without wiring up billing, the cheapest first step is a gateway you already pay through — OpenRouter, Cloudflare or Netlify — or TypeSafe’s own playground, where you can send a question without writing code. The routes are listed on how to use Jev.
The comparison that decides it
Per-token prices are the wrong axis, because the two models are not doing the same amount of work per item. Per job is where the gap shows.
| The job | With Jev | With a large model |
|---|---|---|
| 100,000 viral posts, 14 questions each | 20.4 s, $0.67 | 214 posts for $0.98 on Claude Opus 5, same clock |
| A 26-sheet construction plan set | 2.9 s, $0.0052 | 17 to 21× the cost on their GPT-4.1 pipeline |
| TypeSafe’s four workflow evals | $0.0004 a case, 0.4 s, 67.8% accuracy | $0.1174 a case, 78.1 s, 67.8% on Sonnet 5 |
The eval row is TypeSafe’s own, from their workflow evals. The other two are the builders’. The reasoning behind all three is on Jev vs an LLM.
What makes a Jev bill bigger than expected, and where the cost ends up
Three things, in the order they bite. The state is billed on every call, so sending a whole document to answer one question about its first paragraph is where the money goes; trim the state before you trim the questions. Retries count, and the SDKs retry with backoff when you cross a rate limit. And a two-pass pattern — score everything, then re-ask about the shortlist — doubles the state you send, which is usually still worth it, but it is worth knowing you chose it.
For a ceiling: Will Kelly’s adversarial evaluation ran 123,805 requests across nine experiments and cost $12.69 in total. That is the largest published bill in this directory.
The bigger lesson is where the money goes once the decisions are cheap. Gipp’s priced event desk runs 20,000 events a day through Jev: 14,000 are dropped as noise, 4,700 become logs, 800 earn a draft and 500 go to human review. Only about 4% reaches a model that writes. On his published prices and assumed rates, Jev costs $0.84 of a $29.04 day. They are not invoices, and he says so. His point is that Jev does not end the Jevons effect, it moves it. Watch ten times more sources and the bill grows ten times, but the growth lands in the share of events that escalate to a writer, not in Jev. That share is the number to tune, and a router is how you set it.
Common questions
- Is Jev AI free?
- No. Input is billed at $0.042 per million tokens. Output is free of charge. As of Oct 2026 new signups do not get free credits. TypeSafe reopened signups on 27 Sep 2026 after pausing them for capacity.
- How much does Jev cost per decision?
- Across 15 published runs where the author gave both a volume and a total, the median is $0.000065 per decision — about 15,291 decisions per dollar. The cheapest published run worked out at 170,900 per dollar, the dearest at 5,000.
- Why is output free?
- Because there barely is any. Jev returns a choice, a set of probabilities and a confidence, not generated text, so the response is a few dozen bytes however long the question was. You are paying for the state you send in.
- Does asking more questions cost more?
- Hardly. Every question in a call is answered against the same state, and the state is the part you are billed for. One build asks 61 questions about a single draft post for $0.0004.
- Did Jev 1.14 change the price?
- No. Jev 1.14, released on 14 October 2026, costs the same $0.042 per million input tokens as Jev 1.13, and output is still free. jev-latest moves from 1.13 to 1.14 about a week after the release, around 21 October.
- Do images or long context cost more on Jev?
- No. Both are preview features at the same rate. Images are billed as input tokens with no separate charge per image. The opt-in long-context mode raises the limit to 262k tokens per request and per question at the same price, though TypeSafe says it is slower and less accurate.
- How do I pay for Jev?
- Directly with a TypeSafe API key, or through a gateway that already has your billing: Vercel AI Gateway, OpenRouter, the Cloudflare and Netlify AI gateways, or a LiteLLM proxy. Vercel AI Gateway charges $0.042 per million input tokens and $0 output, billed through AI Gateway. The price per token is the thing to compare; the gateways add their own terms.
Made with Jev is independent and not affiliated with TypeSafe AI. Every figure on this page is the one its author published, linked to where it can be checked.