Jev vs Perplexity Decider: the two cheapest decision models, tested on the same inputs

Perplexity Decider (pplx-decider-v1.2) and Jev take the same request, a state plus typed questions, and return probabilities. Decider is now the cheaper of the two by a wide margin: $0.02 per million input tokens against $0.042, and since v1.2 it reads the state once, like Jev. On the same 196 labelled inputs Decider v1.2 scored 89.6% on Banking77 card intents and 91% on AG News (2026-10-25); Jev scored 86.5% and 91% (2026-10-07). Jev’s advantages are elsewhere: a lower latency and the TypeSafe tooling around it.

Try it on your own text

Edit the example, then sign up free to run it — 15 credits, no card.

Scenario

Route a ticket to the right team, rate its urgency and flag churn risk.

166 / 2,000

Edit freely — the model answers the scenario's questions about whatever you put here.

Model

What Jev actually returned

This is the unedited output of the example above, captured from the live API.

Hi, I've been trying to connect my Stripe account for 3 days and it keeps failing with 'invalid redirect'. I'm losing sales every hour this is down. Please help ASAP.

category

choice

Which team should handle this support ticket?

integrations

  • integrations99%
  • billing1%
  • bug0%
  • account0%

Confidence 99%

urgency

score

How urgent is this ticket?

3.73/4Critical
  • 4 · Critical73%
  • 3 · Urgent27%
  • 0 · Can wait0%
  • 1 · Normal0%
  • 2 · Soon0%

Confidence 77%

churn risk

noul

Is the customer at risk of churning?

Yes80%
Yes 80%No 20%

Captured 2026-10-09 · 0.89 s · jev-1.13.0

At a glance

(vendor) is the company selling the model; (ours) is a call we made through OpenRouter: Jev on 2026-10-07, Decider v1.2 on 2026-10-25.

JevPerplexity Decider
Made byTypeSafePerplexity
Model idjev-latestpplx-decider-v1.2
WeightsClosedv1 open (Apache 2.0); v1.2 not listed
SizeNot disclosed27B (vendor)
List price, input$0.042 / M tokens$0.02 / M tokens
Output tokensFreeFree
Context~32k tokens, shared with the longest question (vendor); 262k opt-in262k tokens (vendor)
ImagesOpt-in preview (vendor), not on JevStationYes
Input tokens, our fit≈ 260 + 0.163 × characters + 43 × questions≈ 48 + 0.163 × characters + 40 × questions
Median latency412 ms (ours)525 ms (ours)
Banking77 card intents86.5% (ours)89.6% (ours)
AG News topics91% (ours)91% (ours)
Credits on JevStation1 / 31 / 3

Credits are standard / large; an evaluation is large above 8,000 characters of state or five questions.

Accuracy and calibration

We sent both models the same 196 labelled inputs, zero-shot, with one line of description per option. The method is on Decision models compared, which runs all the models on the same data. Jev was run on 2026-10-07 and Decider v1.2 on 2026-10-25; the v1 column is the Decider model OpenRouter retired in October, run on 2026-10-07.

TaskJevDecider v1.2Decider v1 (retired)
Banking77, 12 card intents (96)86.5%89.6%86.5%
AG News, 4 topics (100)91%91%93%
Banking77: answers at ≥ 0.9 confidence81% of inputs, 94.9% right73% of inputs, 100% right77% of inputs, 98.6% right
Wrong at ≥ 0.99 confidence (of 196)411
Expected calibration error, AG News0.0860.0400.052

The accuracy is a tie at this sample size. Calibration is where Decider pulls ahead: when it was very sure, it was nearly always right, so a threshold set on its confidence lets fewer mistakes through. v1.2 is a little more cautious than v1, answering fewer inputs above 0.9 but getting all of them right on Banking77. Jev gave a confidence of 1.00 (to two decimals) on 136 of the 196 inputs, which leaves little room to separate easy cases from risky ones.

Both models found a single cancel request hidden at the very end of 20,000 characters of neutral text, so neither truncated the state in our test.

What each call costs

Decider v1.2 is cheaper per token and counts about as many tokens as Jev, so it is cheaper on every request shape we measured. We fitted both on nine requests each (state of 100, 2,000 and 8,000 characters, with one, three and five questions):

RequestJev tokensDecider v1.2 tokensCheaper at list price
1 question, short ticket (~300 chars)~350~140Decider, by about 5×
3 questions, 1,500 characters~640~410Decider, by about 3×
5 questions, 8,000 characters~1,7801,550Decider, by about 2×
12 long questions, 24,000 characters8,9818,310Decider, by about 2×

The retired Decider v1 was different. It read the state once per question, so the five-question request above cost it 6,985 tokens and the twelve-question one 51,882. Pages that told you Decider gets expensive with several questions were describing v1.

On JevStation both models now cost the same credits. The pricing calculator shows the direct cost of each for your own numbers.

When to pick which

  • Lowest cost per call, direct: Decider.
  • Lowest latency: Jev, about 110 ms faster at the median in our runs.
  • You need images judged: Decider. Jev's image input is an opt-in preview that TypeSafe does not call production-ready, and JevStation sends text only.
  • You need a confidence you can act on automatically: Decider was better calibrated in our runs; confirm on your own data.
  • Self-hosting or data residency: Decider v1 weights are open; check whether v1.2 weights are published before relying on it. Jev is API-only.
  • Inputs longer than ~30k tokens: Decider, with a 262k context.

Try both

Run Decider above with a free account (50 credits, no card), or use the side-by-side runner to send the same text and questions to Jev and Decider at once.

Sources

Also see Jev vs Drex Max for Nace.AI’s hosted model on the same System One contract (added 2026-11-20).

JevStation is independent. Jev and Decider are names of their respective owners, TypeSafe and Perplexity; we are not affiliated with either.

Frequently asked questions

Is Perplexity Decider a drop-in replacement for Jev?
For the request format, yes. Through OpenRouter’s System One endpoint or JevStation, Decider accepts the same state and the same Choice, Score and Noul questions and returns answers in the same shape.
Is Perplexity Decider open source?
The first version, Decider v1 27B, was published on Hugging Face under Apache 2.0, and Perplexity said it was fine-tuned from Qwen3.8-27B. OpenRouter describes v1.2 as a new 27B checkpoint but lists no Hugging Face weights for it, so check before you plan to self-host. Jev has no open weights.
Which is more accurate, Jev or Decider?
In our tests they were close. On 96 Banking77 card-intent messages Decider v1.2 scored 89.6% and Jev 86.5%; on 100 AG News articles both scored 91%. Gaps this size are inside the noise for samples of about 100. Decider was better calibrated: at 0.99 confidence or above it was wrong once in 196 inputs, Jev four times.
Did Decider v1.2 change how it bills tokens?
Yes. The v1 model we measured on 2026-10-07 read the state again for every question, so an 8,000-character state with five questions came to 6,985 input tokens. v1.2 reads it once: the same request is about 1,550 tokens, and each extra question adds about 40. OpenRouter has retired v1.
What does Decider cost on JevStation?
The same as Jev: 1 credit for a standard evaluation and 3 for a large one. Until 2026-10-25 it cost 2 and 6, because v1 billed the state once per question. The 50 free credits on sign-up cover both, so you can compare them on this page with a free account.

Related

Make it part of your pipeline

Sign up for 15 free credits, save your own question sets and call the same evaluation from the API.