Jev vs Perplexity Decider: closed API or open weights?

Perplexity's Decider models are the closest thing to a direct price rival: they answer the same kinds of question, they cost less per input token on OpenRouter's list, and the weights are open, so you can also run them yourself. Jev is closed, hosted by TypeSafe, and has been available since 15 September.

What we cannot say is which is more accurate. The numbers that favour Perplexity come from Perplexity's own benchmark; the one independent index we found scored Decider v1.1 below Jev.

Press coverage reports that Perplexity released pplx-decider-v1-27b on 1 October 2026 with open weights under the Apache 2.0 licence, built on Qwen3.8-27B, with a Decisions API that answers yes/no, multiple-choice and scored questions. Version 1.1 followed on 6 October at half the price.

On 16 October, OpenRouter's model list shows Decider V1.2 and Decider Flash V1.2, at $0.02 and $0.01 per million input tokens, both with free output. We have not read Perplexity's own documentation for these versions, so check it before you rely on a price or a limit.

Perplexity says its model scores highest on its own Decision Index benchmark, and one report cites 85.71% for the first version against 84.51% for Jev. These are the company's figures on a test whose method is not fully published, so they are a claim, not a result.

For Jev, independent tests used version 1.13 and showed results that swing with the task: close to the best models on narrow questions, behind on broad ones. Our benchmarks article collects them. Artificial Analysis's independent index is below.

• Jev benchmarks: what do independent tests say?

On Artificial Analysis's Decision Index (read 18 October), Perplexity's Decider v1.1 27B scored 53.1, against 57.5 for Jev 1.13 and 57.1 for Jev 1.14. The index is independent, runs four tests itself and is days old; it listed v1.1 and did not list the newer V1.2 versions.

So on that test Jev scored higher and Perplexity's model costs less. Which matters more depends on your task, and the index is not your data.

Based on what is reported today:

• Choose Jev if you want the hosted service that has been available longest, and its independent tests and documentation, with no machines to run.
• Look at Perplexity's Decider if the lower listed price matters at your volume, or you want open weights you can run, tune and keep in house.
• Test both before you commit. Run the same labelled sample through each and compare accuracy, speed and cost on your own numbers.

More on how Jev compares:

• Jev vs open models: can you skip Jev?
• Jev vs Laya: pay per call or run it yourself?
• Jev cost calculator