Perplexity AI has introduced a new contextual embedding model, pplx-embed-v2-context-9b-preview. The model processes entire documents in a single pass, allowing individual text segments to retain context from the surrounding content.
The company previously launched the pplx-embed-v1 and pplx-embed-context-v1 families on February 26, 2026. On the ConTEB benchmark, the 4B version of the v1 model achieved an nDCG@10 score of 81.96 percent, outperforming models from Voyage and Anthropic.
Internal benchmarks conducted by Perplexity on over 1 billion production pages showed recall improvements compared to Qwen3 embeddings. The v1 models were trained using diffusion-based techniques on approximately 250 billion tokens, followed by multi-stage contrastive learning.
The models support native INT8 and binary quantization, which the company claims provides up to 32x storage savings. API pricing for the v1 lineup ranges from $0.004 to $0.05 per million tokens, depending on the specific variant and size.
