Embeddings

Turn text or supported multimodal inputs into vectors for RAG, semantic search, clustering, and recommendations. Start text workloads with text-embedding-v4; evaluate gemini-embedding-2 when one index must compare supported Google text, image, video, audio, or PDF inputs, and confirm current input types on the model detail page before production use.

Supported models

Model	Dim	Ctx	Price / 1M	Best for
gemini-embedding-2	3072	8K	See model details	Supported Google text and multimodal embedding inputs
gemini-embedding-001	3072	2K	See model details	Text embeddings with task_type support
text-embedding-v4	1024	8K	$0.13	General, English-leaning
tongyi-embedding-vision-plus	Model-specific	Model-specific	See model details	Vision embedding and image-text retrieval
tongyi-embedding-vision-flash	Model-specific	Model-specific	See model details	Lower-latency vision embedding

Example

Python

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.sealink.io/v1",
    api_key=os.environ["SEALINK_API_KEY"],
)

# Single string
res = client.embeddings.create(
    model="text-embedding-v4",
    input="SeaLink helps SEA developers ship AI faster."
)
vec = res.data[0].embedding   # 1024-dimensional vector

# Batch (recommended for performance)
texts = ["Doc 1 content", "Doc 2 content", "Doc 3 content"]
res = client.embeddings.create(model="text-embedding-v4", input=texts)
vectors = [d.embedding for d in res.data]

Parameters

Field	Type	Description
model	string	text-embedding-v4, gemini-embedding-2, gemini-embedding-001, tongyi-embedding-vision-plus, tongyi-embedding-vision-flash
input	string \| string[] \| object	Text to embed; gemini-embedding-2 also accepts supported objects with image, video, audio, pdf, file_uri, mime_type, and optional text.
encoding_format	string	float, base64
dimensions	integer	Optional output dimension override when supported. Gemini embedding models support 128-3072.
output_dimensionality	integer	Alias for dimensions. gemini-embedding-001 and gemini-embedding-2 default to 3072; 768, 1536, and 3072 are recommended.
document_ocr	boolean	Optional PDF OCR control for gemini-embedding-2.
audio_track_extraction	boolean	Optional video audio-track extraction control for gemini-embedding-2.
task_type	string	Use only on models that support task_type, such as gemini-embedding-001. Do not send it to gemini-embedding-2; put the task instruction in input text.

Which one?

gemini-embedding-2: Use when one index must compare text with images, videos, audio, or PDFs. For retrieval, format text queries as task instructions such as task: search result | query: ... . Multiple parts in one gemini-embedding-2 request return one aggregated vector; use separate objects or batch processing when each input needs its own vector. PDF input to gemini-embedding-2 supports one file with up to 6 pages, and one-page files are the safest default. Pricing varies by input type; see the model detail page for current units and prices.
text-embedding-v4: General-purpose text embedding model for RAG, semantic search, and recommendation baselines. It returns 1024-dimensional vectors.
gemini-embedding-001: gemini-embedding-001 remains text-only and supports these task_type values: SEMANTIC_SIMILARITY, CLASSIFICATION, CLUSTERING, RETRIEVAL_DOCUMENT, RETRIEVAL_QUERY, CODE_RETRIEVAL_QUERY, QUESTION_ANSWERING, and FACT_VERIFICATION. Its embedding space is not compatible with gemini-embedding-2, so re-embed stored content before migrating an index.
tongyi-embedding-vision-plus / flash: Use for image, image-text retrieval, and visual similarity tasks. Choose input fields from the model detail page.

Performance tips

Prefer batch input (array form) to reduce request overhead and improve throughput.
If you use cosine similarity, L2-normalize before storing so retrieval can use dot product.
Chunk size: 500-800 tokens with 50-100 token overlap is a safe default.
Before production use, evaluate retrieval quality, reranking, and cost with a representative query set.

Embeddings

Supported models

Model	Dim	Ctx	Price / 1M	Best for
gemini-embedding-2	3072	8K	See model details	Supported Google text and multimodal embedding inputs
gemini-embedding-001	3072	2K	See model details	Text embeddings with task_type support
text-embedding-v4	1024	8K	$0.13	General, English-leaning
tongyi-embedding-vision-plus	Model-specific	Model-specific	See model details	Vision embedding and image-text retrieval
tongyi-embedding-vision-flash	Model-specific	Model-specific	See model details	Lower-latency vision embedding

Example

Python

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.sealink.io/v1",
    api_key=os.environ["SEALINK_API_KEY"],
)

# Single string
res = client.embeddings.create(
    model="text-embedding-v4",
    input="SeaLink helps SEA developers ship AI faster."
)
vec = res.data[0].embedding   # 1024-dimensional vector

# Batch (recommended for performance)
texts = ["Doc 1 content", "Doc 2 content", "Doc 3 content"]
res = client.embeddings.create(model="text-embedding-v4", input=texts)
vectors = [d.embedding for d in res.data]

Parameters

Field	Type	Description
model	string	text-embedding-v4, gemini-embedding-2, gemini-embedding-001, tongyi-embedding-vision-plus, tongyi-embedding-vision-flash
input	string \| string[] \| object	Text to embed; gemini-embedding-2 also accepts supported objects with image, video, audio, pdf, file_uri, mime_type, and optional text.
encoding_format	string	float, base64
dimensions	integer	Optional output dimension override when supported. Gemini embedding models support 128-3072.
output_dimensionality	integer	Alias for dimensions. gemini-embedding-001 and gemini-embedding-2 default to 3072; 768, 1536, and 3072 are recommended.
document_ocr	boolean	Optional PDF OCR control for gemini-embedding-2.
audio_track_extraction	boolean	Optional video audio-track extraction control for gemini-embedding-2.
task_type	string	Use only on models that support task_type, such as gemini-embedding-001. Do not send it to gemini-embedding-2; put the task instruction in input text.

Which one?

gemini-embedding-2: Use when one index must compare text with images, videos, audio, or PDFs. For retrieval, format text queries as task instructions such as task: search result | query: ... . Multiple parts in one gemini-embedding-2 request return one aggregated vector; use separate objects or batch processing when each input needs its own vector. PDF input to gemini-embedding-2 supports one file with up to 6 pages, and one-page files are the safest default. Pricing varies by input type; see the model detail page for current units and prices.

text-embedding-v4: General-purpose text embedding model for RAG, semantic search, and recommendation baselines. It returns 1024-dimensional vectors.

gemini-embedding-001: gemini-embedding-001 remains text-only and supports these task_type values: SEMANTIC_SIMILARITY, CLASSIFICATION, CLUSTERING, RETRIEVAL_DOCUMENT, RETRIEVAL_QUERY, CODE_RETRIEVAL_QUERY, QUESTION_ANSWERING, and FACT_VERIFICATION. Its embedding space is not compatible with gemini-embedding-2, so re-embed stored content before migrating an index.

tongyi-embedding-vision-plus / flash: Use for image, image-text retrieval, and visual similarity tasks. Choose input fields from the model detail page.

Performance tips

Prefer batch input (array form) to reduce request overhead and improve throughput.

If you use cosine similarity, L2-normalize before storing so retrieval can use dot product.

Chunk size: 500-800 tokens with 50-100 token overlap is a safe default.

Before production use, evaluate retrieval quality, reranking, and cost with a representative query set.