Perplexity AI has announced the release of its latest models, the pplx-embed-v2-late, which come in two configurations: a 0.6 billion parameter edge model and a more powerful 9 billion parameter model. These models are designed to cater to various applications, with the edge model optimized for deployment on local devices, while the larger model is intended for creating high-quality indexes.
The performance of these new models is noteworthy, achieving a top score of 92.4% on the MADQA benchmark, which measures the quality of AI-generated answers. However, they also recorded a lower score of 61.2% on the ViDoRe v3 Markdown benchmark, indicating some variability in performance across different tasks.
Both models are released under the MIT license, allowing users to self-host them as needed. This move aligns with the growing trend of making powerful AI tools more accessible for developers and researchers, particularly in edge computing environments.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
