Updated 2026-10-12 · Chad · Topic: Hong Kong forums

Jev Pricing: What It Costs per Post, per 100k and per Million (Measured)

TL;DR: Jev is priced at $0.042 per million input tokens, with output free (TypeSafe docs, as of 2026-10-12). In our runs one post cost between $0.0000214 and $0.0000533 depending on how many questions, how long the criteria and whether we sent context. Our final three-question setup averaged 1,063 input tokens, about $0.000045 per post, or roughly $4.50 per 100,000 posts. Claude Haiku 5.5 cost about $0.000106 per post in the same task.

This post contains Apify affiliate links (no extra cost to you). Costs and comparisons are stated honestly, including where Apify is not the best fit.

Prices below are list prices as of 2026-10-12. Check TypeSafe’s documentation and the console for current pricing before you budget. For what Jev is, see the explainer; for accuracy, see the benchmark.

What is the official Jev price?

Item	Value (TypeSafe docs)
Input tokens	$0.042 per million
Output tokens	Free
Rate limit	80 requests per second and 100K tokens per second per account

Because output is free, your bill is a function of one thing: how many input tokens each request carries.

Why do tokens per request vary?

Each request has two parts:

A fixed part: your question wording and criteria. Every post pays for it again, because each request is independent.
A variable part: the post itself plus any context fields such as opening_post or quoted_post.

For short forum posts the fixed part dominates. In our runs, context fields such as the opening post added roughly 300 tokens. Long, precise criteria cost more, but they also bought accuracy: the intent clarification that added tokens lifted intent accuracy by 2.6 points.

What did it cost in our tests?

Measured on our runs, at $0.042 per million input tokens.

Request shape	Avg input tokens per post	Cost per post	Per 100,000 posts
1 question (sentiment Choice, 4 options); posts average 133 characters	509	$0.0000214	about $2.14
4 questions, short criteria	872	$0.0000366	about $3.66
3 questions, long v3 criteria, thread titles only	959	$0.0000403	about $4.03
3 questions, v2 criteria, plus opening-post context	968	$0.0000407	about $4.07
3 questions, long v3 criteria, replies plus opening-post context (up to 600 characters)	1,269	$0.0000533	about $5.33
Average of the final v3 setup (300 LIHKG posts)	1,063	$0.0000446	about $4.46

How much per 1,000, 100,000 and 1 million posts?

Derived from the table above by simple multiplication (per 1,000 is per post times 1,000; per 1 million is per 100,000 times 10).

Request shape	Per 1,000 posts	Per 100,000	Per 1,000,000
1 question	about $0.021	about $2.14	about $21.40
Final v3 average (3 questions)	about $0.045	about $4.46	about $44.60
3 questions, replies plus context	about $0.053	about $5.33	about $53.30

How does that compare with Claude Haiku 5.5?

Also measured: OpenAI’s Decisions API (gpt-6-luna, $0.10 per million input tokens, output free) averaged 1,011 input tokens per post on the same task, about $0.0001011 per post or $10.11 per 100,000 posts. See Jev vs OpenAI Decisions API.

	Jev (`jev-1.13.0`)	Claude Haiku 5.5
Price	$0.042 per 1M input, output free	$0.10 per 1M input, $0.50 per 1M output
Tokens per post	1,063 input	307 input plus 151 output (batched 15 per request)
Cost per post	about $0.000045	about $0.000106
Per 100,000 posts	about $4.50	about $10.60
Accuracy on 150 held-out LIHKG posts (sentiment)	78.0%	82.7%

Jev was cheaper, Haiku more accurate. Both are tiny numbers next to what labelled data is worth, so cost alone rarely decides it. The benchmark has the full comparison and caveats.

How can you keep Jev costs down?

Trade criteria length against accuracy deliberately. Shorter or fewer questions cost less (509 versus 1,063 tokens in our runs), but our longer, precise criteria raised accuracy. Test both on a labelled sample.
Add context only where it changes the answer. Opening-post context cost roughly 300 tokens and fixed a sarcastic reply, but it is wasted on posts that stand alone.
Pin the model version (jev-1.13.0, not jev-latest) so token counts and tuned thresholds do not move when the alias does.
Route by confidence. In our hybrid, Jev answered the confident rows (51% of sentiment, 83% of relevance, 77% of intent) and Haiku the rest, matching or beating Haiku on all three labels.
Set up the request properly first. See the how-to guide for structured state, thresholds and the 7-step recipe.

Worked example: labelling 1 million forum posts a month

All figures are derived from the measured averages above, not from a separate run.

Item	Jev	Claude Haiku 5.5
Volume	1,000,000 posts	1,000,000 posts
Cost at measured average	about $44.60	about $106
Difference	about $61.40 less with Jev
Rate-limit check	1,000,000 requests over a month is far below 80 requests per second

If you route the uncertain rows to Haiku in a hybrid, your bill sits between the two figures, depending on how many rows Jev handles with confidence. In our test that was 51% to 83% depending on the label.

What is not included here?

Your own tokens may differ: longer criteria, other languages and longer posts all change the count. Measure a sample of your own.
Price is only part of cost. Review time for wrong labels, engineering time and scraping costs are separate.
If you would rather not run this yourself, our LIHKG actor offers the same labels as an optional switch at $0.002 per labelled row with the first 20 rows free per run on the free plan, effective 2026-10-27.

This is independent analysis; neither TypeSafe nor Anthropic endorses or sponsors it. Versions are jev-1.13.0 and claude-haiku-5-5 as of 2026-10-12.

From least effort to most freedom

I run it for you Managed, daily to your sheet You do nothing. From US$299/mo. Use it here Run the free tool Instant, no setup, on this site. Run my actor On Apify, your account Same data, you run it and pay per result. Build your own Sign up for Apify Free account — build the pipeline yourself.

This post contains Apify affiliate links (no extra cost to you). Costs and comparisons are stated honestly, including where Apify is not the best fit.

FAQ

How much does Jev cost? TypeSafe’s docs list $0.042 per million input tokens, and output tokens are free. In our runs a typical three-question forum-post request used about 1,063 input tokens, which is about $0.000045 per post, or about $4.50 per 100,000 posts.

Why does the cost per request vary? Input tokens include a fixed part (your question wording and criteria) and a variable part (the post and any context fields). In our runs the fixed part dominated for short posts, and an opening-post context field added roughly 300 tokens.

Is Jev cheaper than Claude Haiku? In our setup, yes: about $0.000045 per post versus about $0.000106 for Claude Haiku 5.5 batched 15 per request, which is about $4.50 versus $10.60 per 100,000 posts at list prices. Haiku was 3 to 5 points more accurate on our held-out test.

What are the Jev rate limits? TypeSafe’s docs list 80 requests per second and 100K tokens per second per account.

How much would 1 million posts a month cost? At our measured average of 1,063 input tokens per post, about $44.60 for Jev and about $106 for Claude Haiku 5.5. Your own number depends on the length of your criteria and context.

How can I lower my Jev costs? Ask only the questions you need, keep criteria as short as accuracy allows, add context fields only when they change the answer, and route only uncertain rows to a second model. Pin a model version so costs and thresholds do not shift.