Compare
TypeSafe: Jev 1.14 vs Perplexity: Decider V1.2
Both return probabilities. The difference is input, price, and who built the product around that idea.
First, the names
Perplexity has two products that sound like routing. They are not the same thing.
Decisions API / Decider (`pplx-decider-v1.2`, `pplx-decider-flash-v1.2`) returns probabilities instead of text. That is the new one. Docs.
Orchestrator API sends a request to a healthy deployment of a model you already chose. It does not pick the model, and it is not a classifier. Docs.
TypeSafe's product is Jev, also a decision model. Jev Router does pick a model for you. OpenRouter.
How a decision model works
You send a state and named questions. Your code applies a threshold, for example auto-close when p(spam) is above 0.95.
| Type | You ask | You get |
|---|---|---|
| noul | yes/no | probability of yes |
| choice | one of your options | probability per option |
| score | a level on your rubric | probability per level, plus an expected score |
Side by side, as of Oct 18, 2026
Prices below are from the research notes. The spec table further down is the live OpenRouter listing.
| Decider v1.2 | Decider Flash | Jev 1.14 | |
|---|---|---|---|
| Size | 27B | 4B | not disclosed |
| Input price | $0.02 / M | $0.01 / M | $0.042 / M |
| Output price | free | free | free |
| Input | text + images | text + images | text only |
| Context | 256K | 256K | 64K / request, 32K / question |
| Questions | noul, choice, score | same | choice, score, noul, plus Box |
| Weights | v1-27b posted; v1.2 unverified | not verified | closed |
| Launched | API Oct 1, 2026; v1.2 Oct 16 | Oct 16, 2026 | Announced Sep 15; 1.14 on Oct 14 |
Pick Perplexity Decider if
- the input includes images, or is long (up to 256K)
- you want the cheaper per-token option and Flash clears your own eval
- you may want open weights. Only the earlier v1-27b release was confirmed.
Pick TypeSafe Jev if
- you want the vendor that built the stack around decision models and calibrated confidence
- the input is text or JSON and fits in 64K
- you want their SDK and agent skill
Vendor claims to treat carefully
TypeSafe claims "193.6x faster, 444.6x cheaper" on its own workflows and "zero hallucinations." Those are the vendor's numbers.
Perplexity said Decider v1 scores 85.71% across benchmarks, measured through its own API.
For an independent view, see Artificial Analysis's Decision Index.
Related guides: ticket-routingcontent-moderationllm-output-gradingmodel-routinglead-scoring
Live specs
| TypeSafe: Jev 1.14 | Perplexity: Decider V1.2 | |
|---|---|---|
| Type | decision | decision |
| Input | $0.042 / M | $0.020 / M |
| Output | Free / M | Free / M |
| Context | 262K | 262K |
| Inputs | text, image | text, image |
| Tools | no | no |
| Weights | closed | open |
PickTypeSafe: Jev 1.14 if
- the input is text or JSON inside 64K
- you want the lab whose product is the decision model
- you want their SDK
PickPerplexity: Decider V1.2 if
- the input includes images or runs long
- you want the lower listed input price
- you are already on Perplexity's Decisions API
Saved runs
No public playground runs include both models yet.