ModelPerplexityConfirmed
Story 11 of 13 in the Thursday 12 November editionPerplexity releases open-weight Decider v1.3 models with locate spans
Built on Qwen3.8-27B and distilled Qwen3.5-4B backbones under an Apache licence, pplx-decider-v1.3 and pplx-decider-flash-v1.3 add exact character-span extraction across Hugging Face, OpenRouter and the Decisions API.
Edited by Ashutosh Shrivastava
13 Nov 2026, 09:15 IST · 2 min read
Perplexity released open-weight pplx-decider-v1.3 (adapted from Qwen3.8-27B) and KL-distilled pplx-decider-flash-v1.3 (on Qwen3.5-4B) under an Apache licence on 12 November 2026 across Hugging Face, OpenRouter and the Perplexity Decisions API, one day after Google’s Gemini Decisions API. Alongside single-pass Yes/No, multiple-choice and ordered-score probabilities with isolated-question bidirectional attention, v1.3 adds locate, an unmerged LoRA span head returning exact start and end character offsets for NER, PII and extractive QA.
Technical details
- Model backbones and licence
- pplx-decider-v1.3 (Qwen3.8-27B) and pplx-decider-flash-v1.3 (Qwen3.5-4B), Apache licence
- Supported question modes
- Yes/No, multiple choice, ordered score and locate character-offset span extraction
- Extraction and API accuracy
- 69.0 (v1.3) and 63.6 (Flash) across 17 extraction tasks vs 43.3 for Liquid d1; 93.98% judge agreement on 1M API requests
- Distribution channels
- Hugging Face (perplexity-ai/pplx-decider-v1.3), OpenRouter and Perplexity Decisions API
Release details & context
Benchmark margins against Jev 1.14 and internal security log triage
Across Perplexity’s zero-shot evaluations, Decider v1.3 reaches 91.6% mean accuracy over 63 query-classification tasks (+6.54 points over production classifiers), 83.99 nDCG@10 across 72 retrieval suites (+2.24 over TypeSafe AI’s Jev 1.14), 88.16% across 8 languages, 76.99% on five vision document tasks, and flips answers on just 0.27% of reordered multiple-choice prompts versus 2.69% to 4.84% for Jev 1.14. Internally, scanning 1B tokens across 7,000 agent logs in one day surfaced 14 security-rule fixes (following Perplexity’s SPACE sandbox red-teaming) that trimmed alert volume on its busiest rule by 52%.
Sources
- PrimaryPerplexity Hub: Perplexity Decider v1.3: Training, eval, and use casesperplexity.ai
Perplexity release timeline (7 launches tracked)
- Perplexity rolls out updated voice mode for Pro and Max on web and Mac
Perplexity’s updated voice mode runs on OpenAI’s gpt-live-1 and lets Pro and Max users steer multiple Computer sessions and Projects on web and Mac.
- Perplexity Computer launches automated Shopify, Klaviyo and Meta ad workflows
Perplexity launched an e-commerce workflow suite inside Perplexity Computer connecting Shopify, Meta Ads, Klaviyo and 24 other merchant tools.
- Perplexity introduces Teammates personal agent workforce for Max subscribers on the web
Perplexity launched Teammates for Max subscribers on the web, deploying persistent personal agents that coordinate across Computer tasks, memory and Slack.
- Perplexity red-teams 12 frontier models against its SPACE virtual-machine agent sandbox
Perplexity red-teams 12 frontier models on its SPACE VM sandbox, logging zero escapes while general models beat cyber-tuned variants.