PPLX 27B

#31 in Open-Source-Spraakmodelle

Perplexity · 27b · 4× · tolest 28. Aug. 2026

34

Momentum

PPLX 27B is a post-trained language model by Perplexity AI based on Qwen 3.8 27B, optimized specifically for local deployment on NVIDIA DGX Spark as part of Perplexity's "Portable Computer" agent product. It runs entirely locally without cloud connectivity and uses Multi-Token-Prediction technology to accelerate processing. According to Perplexity's own Benchmarks, it significantly outperforms the base Qwen 3.8 27B on agent-based knowledge work tasks. The model weights are available on Hugging Face under a proprietary ("other") license.

Belege (4)