Perplexity pplx-decider v1.1 Beats Jev on Decision Index

What happened

Perplexity has replaced its first open decision model only five days after shipping it. On October 1, 2026, the company released pplx-decider-v1-27b as open weights under Apache 2.0 and launched a hosted Decisions API built on it the same day.12 On October 6, it followed with v1.1.1 The new model card reports an overall score of 61.56 on Decision Index 0.3, up from 56.4 for v1. That puts it ahead of the 57.9 posted by Jev, the TypeSafe AI model that has served as the category's reference point.1

The changelog is short. Perplexity attributes the gain to lifting the causal mask and adding more training data.1 One ranking now lists v1.1 as the strongest open option in the category and describes it as multimodal.1

What a "decider" model actually does

These models are not general chatbots. Perplexity's Decisions API returns answers in three typed formats: yes/no, multiple choice, and an ordered score.2 The aim is to give software a structured verdict it can act on, not prose that a developer has to parse.

The v1 base model is a fine-tune of Qwen3.8-27B with a 262,144-token context window.2 According to Perplexity's documentation, prompts of a few hundred tokens get responses in under two seconds. Inputs near the context ceiling take up to 23 seconds.2 That gap matters for anyone weighing real-time use against long-document review.

Pricing is aggressive. Perplexity charges $0.04 per million input tokens and does not bill for output.2 For comparison, OpenAI's Decisions API, which runs on gpt-6-luna and is in public beta, is listed at $0.10 per million input tokens.1 On input pricing alone, Perplexity costs well under half as much. Free output tokens also help, although typed answers are short, so the output savings may be modest in practice.

Two scoreboards, two stories

The way the leads are measured deserves a closer look. Two different benchmark frameworks are in play, and they tell slightly different stories about the original v1.

At launch, Perplexity compared v1 with Jev on its own 11-benchmark panel of 7,210 samples. There, pplx-decider scored 85.71% to Jev's 84.51%.2 The largest margin was on RAGTruth, a hallucination-detection test, where v1 scored 88.80% against Jev's 77.27%.2 Excluding RAGTruth, the two models look very close.

On Decision Index 0.3, however, the same v1 model scored 56.4, below Jev's 57.9.1 So v1 led Jev on Perplexity's chosen panel and trailed it on the Decision Index. One plausible reading is that v1.1 was built to close that gap. Lifting the causal mask is an architectural change, not a minor tweak.

Both sets of numbers come from Perplexity, through its own panel and its own model card.12 Neither is an independent evaluation. The Decision Index carries a 0.3 version number, which suggests the benchmark itself is still changing.

The ranking puzzle

One October 2026 ranking of decision models adds a further wrinkle. It places Jev first, pplx-decider v1.1 second, and OpenAI's Decisions API third.1 That order holds even though the same ranking cites v1.1's higher Decision Index score.1 It labels v1.1 the "strongest open option" and frames OpenAI's offering as best for single-vendor shops and compliance-focused buyers.1

This suggests the rankers are weighing more than one benchmark figure. Possible factors include track record, ecosystem, reliability, or skepticism toward a model that is only days old. The reasoning behind keeping Jev on top is not spelled out, so readers should treat it as an editorial judgment rather than a measured result.

Why it matters

The pace of iteration stands out. Perplexity shipped a model, an API, and a successor that clearly outscores the original on a headline metric, all in under a week.12 Teams that integrated v1 on launch day now have reason to re-test.

The licensing stands out too. Apache 2.0 open weights on a 27B model mean organizations can self-host a decision engine and avoid sending data to a vendor.2 For regulated industries this matters, and it puts direct pressure on hosted-only competitors.

Our reading is that v1.1 makes Perplexity a serious contender in structured decision models on capability, openness, and price. The claim that it outperforms Jev still rests on self-reported numbers from a young benchmark, and at least one ranking is not yet convinced. Until independent evaluations appear, buyers should run both models against their own workloads before choosing between them on leaderboard scores alone.