Perplexity

Perplexity released Decider v1.3 and Decider Flash v1.3, open-weight decision models that upgrade from label-only classification to returning exact text spans with start and end character offsets via a Locate capability. Across 17 out-of-domain NER, PII and extractive QA tasks, v1.3 averages 69.0 and Flash 63.6, versus 43.3 for Liquid d1 and 33.6 for GLiNER large v2.5; query classification reaches 91.6% over 63 tasks with a 0.27% answer-flip rate. Built on Qwen3.8-27B and Qwen3.5-4B under Apache 2.0, with API input pricing at $0.02 and $0.01 per 1M tokens.

IP protection firm Podqi tested ChatGPT, Gemini, and Perplexity with 6,795 shopping-related prompts across 151 brands and found all three engines cited and surfaced fake brand-impersonating websites, with Perplexity performing worst at 15% of answers containing risky sites. The study exposes systemic gaps in AI search around brand authenticity and e-commerce referral safety; Perplexity said it never claims 100% accuracy.