Pplx Embed
#8perplexity · seit 2026-02-26 · 2× · zuletzt 30. Juni 2026
pplx-embed is a family of multilingual text embedding models released by Perplexity AI on February 26, 2026, comprising four variants: pplx-embed-v1 (standard dense retrieval) and pplx-embed-context-v1 (contextual chunking for RAG), each available at 0.6B and 4B parameter scales. The models are built on Qwen3 base models converted into bidirectional encoders via diffusion-based continued pretraining, and natively produce INT8 or binary-quantized embeddings without requiring instruction prefixes. All weights are available under MIT license on Hugging Face, and the models are also accessible via the Perplexity API. On the MTEB Multilingual v2 retrieval benchmark, pplx-embed-v1-4B (INT8) achieves 69.66% nDCG@10, outperforming Google's gemini-embedding-001 (67.71%).
Fonctionnalités
| Multimodal Capabilities | No multimodal capabilities – text embeddings only (no image, audio, or video input documented) |
| Price per 1M Tokens | pplx-embed-v1-0.6b: $0.004 / 1M tokens; pplx-embed-v1-4b: $0.030 / 1M tokens; pplx-embed-context-v1-0.6b: $0.008 / 1M tokens; pplx-embed-context-v1-4b: $0.050 / 1M tokens |