Perplexity Open-Sources 27B Decider, Edges Jev on 11-Test Panel

Perplexity on October 1, 2026 opened a Decisions API running on pplx-decider-v1-27b, an open-weight 27B model released on Hugging Face under Apache 2.0 and fine-tuned from Qwen3.8-27B. The API answers in three typed shapes (yes/no, multiple choice, and ordered score) at $0.04 per million input tokens with free output. The docs show responses under two seconds for a few hundred input tokens, stretching to 23 seconds near the 262,144-token input ceiling.

Perplexity's own numbers put pplx-decider ahead of TypeSafe's Jev on an eleven-benchmark panel of 7,210 samples: 85.71% overall against Jev's 84.51%, with the biggest gap on RAGTruth (88.80% vs 77.27%) and smaller wins on FinancialPhraseBank, TabFact, and Circa. Jev still leads on four of the eleven tests, including WinoGrande (90.70% vs 83.30%) and BBH (94.27% vs 82.80%). As cellcog.ai noted, "Perplexity's X post cites an 85.71% score across benchmarks; its docs publish no benchmark table", so the figures come from the company, not an independent rerun.

Perplexity landed on a crowded day. explainx.ai reported that Cloudflare also shipped Clef and Clef-flash and AWS's Strands Labs released Strands Decider 2B on October 1, three decider launches aimed at the same classification-and-routing use cases Jev has been serving. It's the eleventh Perplexity story on our tracker in the last ninety days.