pplx-embed-v2-late-0.6b and 9b released
Perplexity · Model release
Source confirmed
Data as of October 7, 2026 · How the score is built
CurrentOpen WeightNon-Reasoning
Released Oct 7, 20264096 tokens contextPerplexity pplx-embed-v2-late-0.6b model card
pplx-embed-v2-late-0.6b
Decision readingpplx-embed-v2-late-0.6b is tracked, but not publicly ranked yet. The profile exposes 2 sourced benchmark rows and leaves unsupported fields blank until a published record exists.
Released Oct 7, 2026 — see all recent releases
pplx-embed-v2-late-0.6b will be repriced, updated or retired. Get each notice with its source and date. Follow model changes
Decision snapshot
Each value carries a field reference instead of floating alone. Markers compare this model with the current ranked and priced catalog; they are not absolute quality thresholds.
Capability
Unranked
field median 52.1Not eligible for a public rank
Price
Self-hosted; infrastructure cost varies
input median $0.95No comparable first-party hosted token rate
Speed
Not measured
field median 97 tok/sTime to first token not measured
Context
4096 tokenstokens
field median 256,000Maximum output length is tracked separately
Strongest published evidence
Published rows are visible, but no category has enough eligible evidence for a comparative rank.
Validate before choosing
2 published rows leave some tracked benchmark slots empty. No comparable first-party API token rate is published.
Source-linked · 2 displayable benchmark rows
Follow model changesBenchmark ledger
External signals opens by default. The marker compares each value with the best source-verified result in the catalog; provisional leaders do not set the reference. Expand the remaining categories for every published row.
External signals2 rows
| Benchmark | Score | Versus best verified row | Gap | Weight | Evidence |
|---|---|---|---|---|---|
| ViDoRe V3 Image (Perplexity report) | Score62.3% | Versus best verified row Best verified: pplx-embed-v2-late-9b · 65.2% | Gap2.9 behind | WeightDisplay only | |
| ViDoRe V3 Markdown (Perplexity report) | Score61.2% | Versus best verified row Best verified: pplx-embed-v2-late-9b · 64.7% | Gap3.5 behind | WeightDisplay only |
Bars run 0–100; the dark tick marks the best source-verified value
All 2 rowsLineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
Radar
Full release historypplx-embed-v2-late-0.6b release history
Radar confirmed these at the source. Use pplx-embed-v2-late-0.6b in your work? Explore Radar to follow supported changes and choose your alerts.
Radar
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- Weights model ID
- perplexity-ai/pplx-embed-v2-late-0.6bPerplexity pplx-embed-v2-late-0.6b model card
- Context window
- 4096 tokensPerplexity Sentence Transformers encoding limits
- Maximum output
- 128-dimensional vector per retained token; no generated textPerplexity pplx-embed-v2-late-0.6b model card
- Knowledge cutoff
- Not sourced yet
- Input modalities
- text, imagePerplexity pplx-embed-v2-late-0.6b model card
- Output modalities
- embeddingPerplexity pplx-embed-v2-late-0.6b model card
- Parameters
- 594M total (checkpoint metadata); 340M active (model card)Hugging Face checkpoint parameter metadata
- Availability
- Hugging Face · Self-hosted inferencePerplexity pplx-embed-v2-late-0.6b model card
- Cloud regions
- Not tracked yet
- Lifecycle
- activePerplexity pplx-embed-v2-late-0.6b model card
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Not documented in the pricing recordPerplexity pplx-embed-v2-late-0.6b model card
- Self-host
- Open weights available; hardware estimate not sourced
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
We track pplx-embed-v2-late-0.6b, but the public leaderboard excludes this profile until enough non-generated benchmark coverage is available. Only published rows appear above.
pplx-embed-v2-late-0.6b is a open weight model with a 4096 tokens context window. No explicit reasoning mode is documented in this profile.
Self-hosted Sentence Transformers MultiVectorEncoder weights. Text and image inputs are supported in separate batches; mixed text-plus-image inputs are not supported. Output is token-level embeddings, not generated text. Queries are capped at 1,024 tokens and documents at 4,096 tokens in the published encoding config.
Perplexity released pplx-embed-v2-late-0.6b on October 7, 2026, with MIT-licensed weights. It is a multilingual text and image retriever built on Qwen3.5 with bidirectional attention. The checkpoint has 594M total parameters in Hugging Face metadata; the model card reports 340M active parameters. Its Sentence Transformers config limits queries to 1,024 tokens and documents to 4,096 tokens; the trunk positional limit of 262,144 is not the published encoding limit. Encode text and images in separate batches; mixed text-plus-image inputs are not supported. Perplexity reports public ViDoRe V3 nDCG@10 on a 0–100 scale. Image and Markdown retrieval use separate input representations. These provider-reported results remain display-only and do not establish general LLM capability scores. The two models share a multi-vector embedding space; each retained token has a 128-dimensional vector, and query–document pairs use MaxSim scoring.
pplx-embed-v2-late-0.6b sits in the Perplexity Embed v2 Late family with pplx-embed-v2-late-9b. 2 of 665 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Last updated October 7, 2026. Runtime fields remain blank until a sourced snapshot exists.
Deployment options
Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.
Published weights are available, but BenchLM does not yet have a sourced parameter and VRAM profile for this exact model. Hardware cost estimates stay unavailable until that sizing record is complete.
Estimate VRAM from known parametersQuestions
How does pplx-embed-v2-late-0.6b perform overall in AI benchmarks?
pplx-embed-v2-late-0.6b has 2 source-displayable benchmark rows, but it does not qualify for a public overall rank. The available rows remain visible by category without being converted into a site-wide score. Missing evidence stays blank instead of being estimated from an earlier model.
Is pplx-embed-v2-late-0.6b open source?
pplx-embed-v2-late-0.6b is an open-weight model from Perplexity. Its weights can be downloaded for local or hosted deployment, subject to the published license. Open weight does not automatically mean open source: training data and training code may remain private, and commercial restrictions can still apply.
Which sibling models are related to pplx-embed-v2-late-0.6b?
pplx-embed-v2-late-0.6b belongs to the Perplexity Embed v2 Late family. Related tracked variants include pplx-embed-v2-late-9b. A sibling link indicates shared lineage or a documented configuration relationship; it does not mean the variants have identical pricing, context limits, benchmark evidence, or deployment behavior. Compare before switching.
Does pplx-embed-v2-late-0.6b have full benchmark coverage on BenchLM?
No. pplx-embed-v2-late-0.6b currently has 2 source-displayable rows across 665 tracked benchmark slots. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of pplx-embed-v2-late-0.6b?
pplx-embed-v2-late-0.6b has a documented context window of 4096 tokens. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.