pplx-embed-v1-0.6b

Name: pplx-embed-v1-0.6b
Author: perplexity-ai

by perplexity-ai

Diffusion-pretrained 0.6B text embeddings with INT8 quantization: SOTA at sub-1B scale

120Kdl/month

0.6Bparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

perplexity-ai/pplx-embed-v1-0.6b

Feature URI

mixpeek://text_extractor@v1/perplexity_pplx_embed_v1_06b

Overview

pplx-embed-v1-0.6B is Perplexity AI's lightweight text embedding model built on diffusion continued pre-trained Qwen3 with bidirectional attention. It natively produces INT8-quantized embeddings, reducing storage requirements by 4x compared to FP32 while maintaining retrieval quality. At just 0.6B parameters, it achieves 68.6 nDCG@10 on MTEB Retrieval, beating the much larger Qwen3-Embed-0.6B (61.2) and BGE-M3 (62.3).

The model supports 32K context length and 1024-dimensional embeddings, with optional binary quantization for 32x storage reduction. On Mixpeek, pplx-embed provides a fast, storage-efficient embedding backbone for text-heavy retrieval pipelines where index size and inference cost are primary constraints.

Architecture

Bidirectional attention transformer built on diffusion continued pre-trained Qwen3. 0.6B parameters. 32K context length. Natively produces INT8-quantized 1024-dimensional embeddings. Supports binary quantization for 32x storage reduction.

Mixpeek SDK Integration

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so pplx-embed-v1-0.6b runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns_your_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection_id: "col_your_collection",
      documents: [
        {
          document_id: "asset-00412",
          // The vector name has to match a vector index on the collection.
          vectors: { "text-embedding": yourVector },
          payload: { source_key: "archive/2026/asset-00412" },
        },
      ],
    }),
  },
);

// Managed alternative, if this exact model is not the requirement:
// text_extractor@v1 runs intfloat/multilingual-e5-large-instruct
// (1024-d) over a bucket, with no inference of your own.

Capabilities

68.6 nDCG@10 on MTEB Retrieval: SOTA at sub-1B scale
Native INT8 quantization (4x storage reduction)
Optional binary embeddings (32x storage reduction)
32K context window for long documents
Beats BGE-M3 and Qwen3-Embed-0.6B on retrieval benchmarks

Use Cases on Mixpeek

Storage-efficient text indexing: embed large document collections with minimal disk footprint

High-throughput RAG pipelines: fast embedding generation for retrieval-augmented generation

Cost-sensitive text search: strong retrieval quality at minimal compute and storage cost

Benchmarks

Dataset	Metric	Score	Source
MTEB Retrieval (en)	nDCG@10	68.6	Perplexity AI, 2026: arxiv,2602.11151
BERGEN End-to-End RAG	Avg score	Beats Qwen3-embedding-4B on 3/5 tasks	Perplexity AI, 2026: arxiv,2602.11151