Perplexity: Embed V1 0.6B

Perplexity: Embed V1 0.6B is a text embedding model from Perplexity, released on 16 March 2026. It belongs to the pplx-embed-v1 family, which Perplexity describes as state of the art text embedding models built for real-world, web-scale retrieval. The 0.6B parameter variant targets lightweight, low-latency use, making it the smaller option in the family rather than the largest.

The model takes text as input and returns embeddings as output, so it does not generate prose or chat replies. It is optimized for standard dense text retrieval, which is the common pattern behind semantic search and similarity matching. The context window is 32,000 tokens, and the maximum output is listed as 28,800. Only text input is supported, and no request parameters beyond the basics are documented for this model.

Because the model produces dense embeddings for retrieval, the natural use cases are semantic search over document collections, retrieval augmented generation pipelines where you need to fetch relevant passages before passing them to a chat model, deduplication or clustering of text, and recommendation or matching systems that rank items by meaning rather than keyword overlap. The 0.6B size and low-latency focus suit workloads where you embed large volumes of text, such as indexing a web-scale corpus or handling high query throughput, and the 32,000 token context means you can embed fairly long documents in a single request instead of splitting them into small chunks.

You call the model through api14's OpenAI-compatible API using the model id "perplexity/pplx-embed-v1-0.6b". Because the interface follows the OpenAI embeddings format, you send your text as input and receive a vector back, and existing OpenAI client libraries work without modification once you point them at the api14 base URL and set that model id. The supported parameters list for this model is empty, so plan on sending only the required fields and avoid relying on optional knobs such as dimensions or encoding format overrides.

If your task is text generation, summarization, or conversation rather than retrieval, an embeddings model is the wrong tool and a chat or completion model will fit better. Similarly, if you need multimodal embeddings over images or audio, this model only accepts text. Within retrieval work, the main trade-off is the 0.6B size: it is built for lightweight, low-latency serving, so if you find retrieval quality insufficient on a demanding corpus, a larger embedding model may be worth considering.