Perplexity's New Decision Models Take No 2 Spot on HF Index

These multimodal models use increased training compute and larger datasets to improve calibration and robustness in vision and text tasks. Perplexity added reranking capabilities and retrained the models to be inference-aware to enhance overall efficiency. To facilitate rapid iterations, the company introduced model slugs pplx-decider-latest and pplx-decider-flash-latest that point to the most recent checkpoints to prevent regression in capabilities.