Jev vs Perplexity Decider: which decision model to use
Facts checked 2026-10-23 on TypeSafe's and Perplexity's own docs and posts. “not verified” means we could not confirm it. We did not run either model.
Jev or Perplexity Decider for typed, probability-based decisions? Both are decision models: you send content as state plus typed questions (yes/no, choice, score) and get probabilities back instead of text. Perplexity's pplx-decider-flash-v1.2 is a 4B model that can now run on an Apple silicon Mac through the Decider for Mac app with no API key or usage bill, and its hosted API lists $0.01 per million input tokens. TypeSafe's Jev 1.14 is a hosted API at $0.042 per million input tokens with a 64k per-request context; text is its production input and images are an early opt-in feature.
What's new
On 22 Oct 2026 Perplexity's developer account announced Decider for Mac, a menu-bar app that runs pplx-decider-flash-v1.2 on the Mac and serves the same Decisions API locally, without an API key. You ask a question about text or an image and get a probability back. Perplexity's CEO described it as running the decision model fully locally, without data leaving the machine.
| Jev (TypeSafe) | Perplexity Decider | |
|---|---|---|
| Made by | TypeSafe | Perplexity |
| Current models | jev-1.14.0 (from 14 Oct 2026), jev-1.13.0 | pplx-decider-v1.2 (hosted), pplx-decider-flash-v1.2 (hosted and on-device) |
| Model size | Not published by TypeSafe | Flash: 4B (Perplexity). Larger model: not verified |
| Question types | Choice (one of your options), Score (ordered rubric), Noul (yes/no probability); Box for image locations (early) | noul (yes/no), choice (one of your options), score (level on your rubric) |
| Input | Text, JSON or arrays of text; images and Box behind an opt-in header, labelled not yet production-ready | Text, JSON or images |
| Hosted price | $0.042 per million input tokens; output free | $0.01 (flash) and $0.02 (v1.2) per million input tokens; output free |
| Context length | 64k per request / 32k per question; 262k with an opt-in long-context header | not verified |
| Run on your own machine | No on-device option found in TypeSafe's docs | Decider for Mac: Apple silicon (M1+), macOS 13+, 16 GB memory, ~6 GB disk; downloads a 5.8 GB q8_0 GGUF build |
| API key needed | Yes (hosted API) | Hosted: yes. Decider for Mac: no, serves http://127.0.0.1:8080/v1/decisions |
| Weights | Not published | GGUF build fetched from Hugging Face repo perplexity-ai/pplx-decider-flash-gguf-v1.2; licence not verified |
| Independent benchmarks | not verified (we ran none) | not verified (we ran none) |
How we compare: the two models are listed in alphabetical order and get the same fields, taken from each vendor's own docs on the checked date. We did not benchmark either model and give no scores. Nobody paid for this page. How we review.
Verdict: when to use which
Use Jev if you want a hosted decision API with documented long-context and calibration guidance, and you can send your content to a third-party API.
Use Perplexity Decider if content must stay on your machine (Decider for Mac reads it locally), you want image input in production use, or you want the lower listed hosted price; the same request moves to the hosted API by changing the URL and adding a key.
Before either one: if the answer can be computed with a plain rule, use code. If the task is stable and high-volume with good labels, a fine-tuned classifier is worth benchmarking too. Neither model writes text, so keep a chat model where the output must be prose or tool arguments.
What we don't know yet
- How Jev and pplx-decider models compare on accuracy or calibration; we found no head-to-head results.
- The size of pplx-decider-v1.2 and of Jev, and the context limits of Perplexity's models.
- The licence of the downloadable Decider flash weights.
- How quickly Decider for Mac answers on a given Mac.
FAQ
What is a decision model?
A model that reads natural-language content and answers typed questions with probabilities instead of writing text. TypeSafe calls Jev a "System One" model; Perplexity calls its Decider models decision models. Neither writes replies, code or explanations.
Is Jev open source or downloadable?
We found no downloadable weights or on-device option in TypeSafe's docs. Jev is served through TypeSafe's hosted endpoint, POST /v1/systemone.
Is Decider for Mac free?
Perplexity says there is no API key and no usage bill when you run it on your Mac. You need an Apple silicon Mac, and the app downloads the model (about 5.8 GB) on first launch; it may ask for a Hugging Face read token.
Does local Decider give the same answers as the hosted API?
Not guaranteed. Perplexity says the Mac weights are a q8_0 GGUF build, not the hosted deployment, and advises re-checking thresholds against the hosted model before production.
Which is more accurate?
Unknown. We have not seen a head-to-head benchmark from either vendor or an independent party, and we did not test either model. Measure both on your own labelled data.
Sources
Related reading
- Another set of new assistants compared: Claude Duet vs Meta Muse vs OpenAI Dots