Published on Hugging Face on October 1, 2026 as perplexity-ai/pplx-decider-v1-27b. It is a Qwen3.8-27B fine-tune with a decision readout, choice and noul (yes/no) question types, optional image inputs, and a 262,144-token position limit. Perplexity's 11-benchmark API-measured panel lists 85.71% overall versus Jev 84.51% and the Qwen base 74.76%; Jev still leads several hard rows including WinoGrande, BBH, and JevBench public hard. Local inference wants Python 3.12+ and about 49 GiB of GPU memory.