pplx-embed-v2-late-0.6b

Released
2026-10-07
Last refreshed
2026-10-08
Status
Researched today
Open sourceCommercial use: permittedMultimodal

pplx-embed-v2-late-0.6b is available now for general LLM work with open-source; evaluate it while provider pricing coverage matures.

Perplexity Labs releases · 5 in the last 12 months · this family litChangelog →
Specifications
Released
2026-10-07
Parameters
0.6B (340M active)
Specialization
embedding
Openness
Open source
License
MITOSI-approvedCommercial use: permitted
Weights
Available
Code
Unknown
Training
Fine-tuned
Created by
Perplexity Labs

Developing AI for complex problem-solving.

San Francisco, California, United States
Founded 2022
Website
Pricing

No tracked provider token pricing is available yet.

Links
Website HuggingFace

About

pplx-embed-v2-late-0.6b is Perplexity's open multimodal late-interaction (ColBERT-style) embedding model in the pplx-embed-v2-late family, built on Qwen3.5 with bidirectional attention. It emits one 128-dimensional vector per token and scores query–document similarity with MaxSim. The 0.6B and 9B checkpoints share an embedding space (0.6B can query a 9B index). Text and images/visual documents are supported; mixed text+image batches are not. Released 2026-10-07 under MIT. Available as open weights on Hugging Face; Perplexity's blog says a hosted API is forthcoming (HF safetensors total ~594M params). Do not confuse with pplx-embed-v2-context-9b-preview.

Provider price ladder

No tracked provider token pricing is available for this model yet.

Capabilities

VisionMultimodal

Benchmark peer barsfor Coding

No task-mapped benchmark peers are available for this model yet.

Migration checks

No linked migration route is available for this model yet.

API versions

pplx-embed-v2-late-0.6bperplexity-ai/pplx-embed-v2-late-0.6b