pplx-embed-v1-4B is Perplexity’s larger independent text-embedding edition in the v1 family. It represents queries or documents as 2560-dimensional vectors for semantic search and retrieval. The manufacturer builds the family on Qwen3 with diffusion continued pretraining and uses mean pooling for the text representation. This independent encoder is distinct from pplx-embed-context-v1, which represents document chunks using surrounding context.
The model does not require an instruction string or a query/passage prefix. Prepare the text you want represented rather than adding a role convention borrowed from another family. A search collection can contain source documents, focused passages or short text records. Keep each item’s original wording and identifier beside its vector so a retrieved result remains readable and traceable.
A practical task is to retrieve a section of a technical explanation for a question about its conditions. Preserve the section heading, the rule and the exception together. The query wording is embedded independently, then compared with the prepared corpus representations. A later RAG stage uses the returned passage text, not the vector itself, as its source material.
The maker explicitly states that the native embeddings are unnormalised and INT8-quantised. Compare them with cosine similarity. A raw dot product on these unnormalised values is not the documented equivalent of cosine comparison. The representation type and similarity function therefore belong together when choosing how a search index stores and compares the output.
The model card also documents binary output and marks Matryoshka representation support. Binary components and their packed storage are different from ordinary floating-point arrays, so the index must interpret the selected representation. The full embedding dimension describes the representation’s components, not the number of bytes in every encoding. Keep the dimension and output type consistent across the query and document sides.
The model’s published context is 32K. That allows substantial text to be represented, while passage granularity should still follow what the search is meant to return. A complete section can retain a definition and its qualifications; a focused excerpt can point directly to an instruction. Oversized material should be prepared around the chosen input window rather than assuming its omitted tail remains represented.
Preserve a parent-document identifier and source version outside the vector. If the text changes, update the corresponding corpus representation and retain the connection to the new source. For a collection containing multiple languages, inspect the actual questions and candidate passages in the intended language mix rather than inferring equal relevance from a multilingual label alone.
The v1-4B model embeds each supplied text independently. Adding several strings to a batch does not turn it into the separate context-v1 chunk encoder. When the surrounding document must shape each chunk’s representation, that task belongs to the contextual edition and its own preparation.
The smaller pplx-embed-v1-0.6B returns 1024 dimensions, while this 4B edition returns 2560. Keep their exact identities and index representations separate unless a documented compatibility path is established. The useful result here is an independent query or source-text vector, with cosine comparison and the selected quantised representation recorded for the retrieval task.
", "aboutText": "pplx-embed-v1-4B is Perplexity’s larger independent text-embedding edition in the v1 family. It represents queries or documents as 2560-dimensional vectors for semantic search and retrieval. The manufacturer builds the family on Qwen3 with diffusion continued pretraining and uses mean pooling for the text representation. This independent encoder is distinct from pplx-embed-context-v1, which represents document chunks using surrounding context.\n\n### Embed the actual text directly\n\nThe model does not require an instruction string or a query/passage prefix. Prepare the text you want represented rather than adding a role convention borrowed from another family. A search collection can contain source documents, focused passages or short text records. Keep each item’s original wording and identifier beside its vector so a retrieved result remains readable and traceable.\n\nA practical task is to retrieve a section of a technical explanation for a question about its conditions. Preserve the section heading, the rule and the exception together. The query wording is embedded independently, then compared with the prepared corpus representations. A later RAG stage uses the returned passage text, not the vector itself, as its source material.\n\n### Native representation and comparison\n\nThe maker explicitly states that the native embeddings are unnormalised and INT8-quantised. Compare them with cosine similarity. A raw dot product on these unnormalised values is not the documented equivalent of cosine comparison. The representation type and similarity function therefore belong together when choosing how a search index stores and compares the output.\n\nThe model card also documents binary output and marks Matryoshka representation support. Binary components and their packed storage are different from ordinary floating-point arrays, so the index must interpret the selected representation. The full embedding dimension describes the representation’s components, not the number of bytes in every encoding. Keep the dimension and output type consistent across the query and document sides.\n\n### Length and document boundaries\n\nThe model’s published context is 32K. That allows substantial text to be represented, while passage granularity should still follow what the search is meant to return. A complete section can retain a definition and its qualifications; a focused excerpt can point directly to an instruction. Oversized material should be prepared around the chosen input window rather than assuming its omitted tail remains represented.\n\nPreserve a parent-document identifier and source version outside the vector. If the text changes, update the corresponding corpus representation and retain the connection to the new source. For a collection containing multiple languages, inspect the actual questions and candidate passages in the intended language mix rather than inferring equal relevance from a multilingual label alone.\n\n### Independent and contextual editions\n\nThe v1-4B model embeds each supplied text independently. Adding several strings to a batch does not turn it into the separate context-v1 chunk encoder. When the surrounding document must shape each chunk’s representation, that task belongs to the contextual edition and its own preparation.\n\nThe smaller pplx-embed-v1-0.6B returns 1024 dimensions, while this 4B edition returns 2560. Keep their exact identities and index representations separate unless a documented compatibility path is established. The useful result here is an independent query or source-text vector, with cosine comparison and the selected quantised representation recorded for the retrieval task.", "editorialOverlay": { "copy_fields": null, "copy_sha256": null, "copy_source": null, "html_sha256": "7c3c8b79f45b602e8c472835e123e5b2d7a00b5a4747927a075788271d7524ff", "module_sha256": "ee9dc1c50b1e6a8fb7fa062030fcb4646b9aaaeb08be513b723a5a92a7290f06", "raw_model_unchanged": true, "schema": "ailoot-native-model-editorial-v1" }, "locale": "en", "methods": [ { "html": "perplexity/pplx-embed-v1-4bPerplexity pplx-embed-v1-4B encodes independent text into 2560-dimensional retrieval vectors with a 32K context. It requires no instruction prefix and natively produces unnormalised INT8 embeddings, compared with cosine similarity.
This description covers the model’s capabilities. Available in Ailoot inputs, outputs and options are listed for the selected method below.
pplx-embed-v1-4B is Perplexity’s larger independent text-embedding edition in the v1 family. It represents queries or documents as 2560-dimensional vectors for semantic search and retrieval. The manufacturer builds the family on Qwen3 with diffusion continued pretraining and uses mean pooling for the text representation. This independent encoder is distinct from pplx-embed-context-v1, which represents document chunks using surrounding context.
The model does not require an instruction string or a query/passage prefix. Prepare the text you want represented rather than adding a role convention borrowed from another family. A search collection can contain source documents, focused passages or short text records. Keep each item’s original wording and identifier beside its vector so a retrieved result remains readable and traceable.
A practical task is to retrieve a section of a technical explanation for a question about its conditions. Preserve the section heading, the rule and the exception together. The query wording is embedded independently, then compared with the prepared corpus representations. A later RAG stage uses the returned passage text, not the vector itself, as its source material.
The maker explicitly states that the native embeddings are unnormalised and INT8-quantised. Compare them with cosine similarity. A raw dot product on these unnormalised values is not the documented equivalent of cosine comparison. The representation type and similarity function therefore belong together when choosing how a search index stores and compares the output.
The model card also documents binary output and marks Matryoshka representation support. Binary components and their packed storage are different from ordinary floating-point arrays, so the index must interpret the selected representation. The full embedding dimension describes the representation’s components, not the number of bytes in every encoding. Keep the dimension and output type consistent across the query and document sides.
The model’s published context is 32K. That allows substantial text to be represented, while passage granularity should still follow what the search is meant to return. A complete section can retain a definition and its qualifications; a focused excerpt can point directly to an instruction. Oversized material should be prepared around the chosen input window rather than assuming its omitted tail remains represented.
Preserve a parent-document identifier and source version outside the vector. If the text changes, update the corresponding corpus representation and retain the connection to the new source. For a collection containing multiple languages, inspect the actual questions and candidate passages in the intended language mix rather than inferring equal relevance from a multilingual label alone.
The v1-4B model embeds each supplied text independently. Adding several strings to a batch does not turn it into the separate context-v1 chunk encoder. When the surrounding document must shape each chunk’s representation, that task belongs to the contextual edition and its own preparation.
The smaller pplx-embed-v1-0.6B returns 1024 dimensions, while this 4B edition returns 2560. Keep their exact identities and index representations separate unless a documented compatibility path is established. The useful result here is an independent query or source-text vector, with cosine comparison and the selected quantised representation recorded for the retrieval task.
POST /v1/embeddingstext, token_idsembedding| Field | Type | Requirement | Values & limits | Default |
|---|---|---|---|---|
Fieldmodel | Typestring | RequirementRequired | Values & limits perplexity/pplx-embed-v1-4b | Default— |
Fieldinput | Typestring|string[]|integer[]|integer[][] | RequirementRequired | Values & limits Minimum items: 1 Conditions: Nonempty homogeneous batch. Token IDs are nonnegative integers. Input budget: Maximum: 32,000 · Text unit: UTF8 bytes · Token unit: token-array elements | Default— |
Fieldencoding_format | Typestring | RequirementOptional | Values & limits float | Defaultfloat |
Fielddimensions | Typeinteger | RequirementNot supported | Values & limits Minimum: 1 | Default— |
Fielduser | Typejson_value | RequirementOptional | Values & limits Conditions: Source allows this JSON metadata field; exact upstream semantics are model-specific. | Default— |
Fieldsession_id | Typejson_value | RequirementOptional | Values & limits Conditions: Source allows this JSON metadata field; exact upstream semantics are model-specific. | Default— |
Fieldtrace | Typejson_value | RequirementOptional | Values & limits Conditions: Source allows this JSON metadata field; exact upstream semantics are model-specific. | Default— |
Fieldprovider | Typeobject | RequirementOptional | Values & limits Conditions: provider.options rejected | Default— |
text_to_embedding
| Parameter | Contract |
|---|---|
Parametercontent_type | Contractapplication/json |
Parameterrequired | Contractmodel, input |
Parameterinput_shapes | Contractstring, string[], nonnegative_integer_token_array, nonnegative_integer_token_array[] |
Parameterencoding_format | Contractfloat |
Parameterdimensions_override | Contractfalse |
Parametermax_input_budget_units | Contract32000 |
Parametermultimodal_objects | Contractfalse |
Parametermaterial_types | Contracttext, token_ids |
Lifecycle: synchronous_json
Keep every vector index and dimension.
Multimodal model name does not admit image objects on this exact path.
Tariff references do not add request options. Supported parameters are listed for the selected method.
| Configuration | Rate schedule | Charge |
|---|---|---|
| ConfigurationBase token rates | Rate scheduleBase rate | Charge Input · Input tokens $0.00006 / 1K tokens Output · Output tokens $0 / 1K tokens Input · Cache read — / 1K tokens Input · Cache write — / 1K tokens Input · Cache write, 1 hour — / 1K tokens |
Ailoot reserves an estimated amount before a request. After the result, it settles the final charge from measured usage. Streaming settlement uses the final usage record. How reservation and settlement work →
set -eu\n: "${AILOOT_API_KEY:?Set your Ailoot API key}"\npython3 - <<'PY_REQUEST'\nimport json,os\nfrom pathlib import Path\np=json.loads('{"model": "perplexity/pplx-embed-v1-4b", "input": ["First document", "Second document"]}')\n\nPath("request.json").write_text(json.dumps(p,ensure_ascii=False))\nPY_REQUEST\ncurl --fail-with-body --silent --show-error https://api.ailoot.ai/v1/embeddings \\\n -H "Authorization: Bearer $AILOOT_API_KEY" -H 'Content-Type: application/json' \\\n --data-binary @request.json -D response-headers.txt -o result.json\npython3 - <<'PY_RESULT'\nimport json\nfrom pathlib import Path\nr=json.loads(Path("result.json").read_text())\nfor x in r.get("data",[]): print(json.dumps({"index":x.get("index"),"embedding":x.get("embedding")}))\nPY_RESULT\nfrom openai import OpenAI\nimport os\n\nclient=OpenAI(base_url="https://api.ailoot.ai/v1",api_key=os.environ["AILOOT_API_KEY"],max_retries=0)\nresult=client.embeddings.create(model='perplexity/pplx-embed-v1-4b',input=["First document","Second document"])\nprint(result)\nfor item in result.data:print(item.index,item.embedding)\njson
Keep every complete returned response, file and content part.
Keep every vector with its index. Vector count must equal input count; dimensions must be consistent.
No embedding job endpoint is introduced.
| Field | Type |
|---|---|
Fieldobject | Typestring |
Fieldmodel | Typestring |
Fielddata[].index | Typeinteger |
Fielddata[].embedding | Typefinite float[] |
Fieldusage | Typeobject |
| Response | Condition | Next step |
|---|---|---|
Response400 · invalid_request_error | ConditionInvalid required fields, material representation or parameter value | Next stepCorrect the indicated fields before a new creation request. |
Ailoot Ultra perplexity/pplx-embed-v1-4b
Use the method paths shown above.