Skip to content
Model infoModelsSnapshot: 15 Oct 2026

ByteDance / Skylark Embedding

Skylark Embedding Vision (251215)

Skylark Vision embedding revision 251215 for multimodal retrieval.

Embeddings
availableClosed weightsEmbeddings
Official documentation (opens in a new tab)Reviewed 12 Oct 2026
6 of 18 specification fields collected
Input token limit128KSee specification notes below
Vector dimensions1024, 2048See specification notes below
Input billingSee ratesBytePlus ModelArk
Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
128KSource for Context / input limit (opens in a new tab)

Provider K-token notation retained.

Maximum output
Not collected
Input modalities
Text, Image, VideoSource for Input modalities (opens in a new tab)
Output modalities
EmbeddingsSource for Output modalities (opens in a new tab)
Reasoning
Not collected
Tool calling
Not collected
Structured output
Not collected
Streaming
Not collected
Batch processing
Not collected
Fine-tuning
Not collected
Knowledge cutoff
Not collected
Release date
Not collected
Parameters
Not collected
Architecture
Not collected
Languages
Chinese, EnglishSource for Languages (opens in a new tab)
Context caching
Not collected
Embedding dimensions
1024, 2048Source for Embedding dimensions (opens in a new tab)

Dense vector output; default 2048.

Published request quota
1.2K Max RPM; 1200K Max TPMSource for Published request quota (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

BytePlus ModelArk · standard

Provider source (opens in a new tab)
API model ID
skylark-embedding-vision-251215
Access
ModelArk API access requires an eligible BytePlus account and enabled model. Published quotas are account-dependent.
Regions
Not collected
Service tier
standardSource for Service tier (opens in a new tab)
API endpoint URL
https://ark.ap-southeast.bytepluses.com/api/v3/embeddings/multimodalSource for API endpoint URL (opens in a new tab)
Input modalities
Text, Image, VideoSource for Input modalities (opens in a new tab)
Embedding dimensions
1024, 2048Source for Embedding dimensions (opens in a new tab)
Source video limits
MP4/AVI/MOV; at most 50 MB per video; no audio understanding.Source for Source video limits (opens in a new tab)
Image input limits
JPEG, PNG, WebP, BMP, TIFF, ICO, DIB, ICNS, SGI or JPEG2000; aspect ratio 1:100–100:1; at most 36 million pixels.Source for Image input limits (opens in a new tab)
Configuration discrepancy
Model catalog gives a 128K context; this API page separately limits each text to 8,000 tokens and 100,000 UTF-8 bytes. Preserve both scopes.Source for Configuration discrepancy (opens in a new tab)
Multimodal fusion
Text, image and video items are fused into one vector; no fixed item-count cap stated.Source for Multimodal fusion (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 12 Oct 2026
CategoryUsageRateUnit
EmbeddingInference input (image)1$0.3251M tokens
EmbeddingInference input (text)1$0.1251M tokens

Rate conditions

  1. Provider publishes a Skylark embedding-vision family rate. Linked to the current 251215 catalog release; video-specific billing and retired-release rates are not established.
Pricing source (opens in a new tab)

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Volcengine Ark

Provider source (opens in a new tab)
API model ID
doubao-embedding-vision-251215
Access
Current native model; activation and an Ark API key required. No inference executed.
Regions
cn-beijing
API endpoint URL
https://ark.cn-beijing.volces.com/api/v3/embeddings/multimodalSource for API endpoint URL (opens in a new tab)
Input modalities
Text, Image, VideoSource for Input modalities (opens in a new tab)
Context / input limit
128KSource for Context / input limit (opens in a new tab)
Embedding dimensions
1024, 2048Source for Embedding dimensions (opens in a new tab)
Multimodal fusion
Mixed input items are vectorized as one combined sample; unlimited item count is still bounded by context and media limits.Source for Multimodal fusion (opens in a new tab)
Embedding output modes
Dense; optional sparse for text-only input; optional multi-vector for text/image/video. Sparse and multi-vector switches are independent and default disabled.Source for Embedding output modes (opens in a new tab)
Output formats
float, base64, nullSource for Output formats (opens in a new tab)

Dense-vector encoding format; not generated text or a JSON-schema feature.

Source video limits
MP4/AVI/MOV, lowercase extension; at most 50 MB per video; video audio is not interpreted.Source for Source video limits (opens in a new tab)
Published request quota
Maximum 15,000 requests/minute and 1,200,000 tokens/minute; load and access method may reduce throughput.Source for Published request quota (opens in a new tab)
Output billing
Text and image input rates are separate. Image accounting: min(width × height / 784, 1312) input units in the published tariff.Source for Output billing (opens in a new tab)
Endpoint availability
Listed in current native catalogSource for Endpoint availability (opens in a new tab)
Embedding task instructions
instructions supports task-specific query/corpus templates; the guide recommends configuring it for the workload.Source for Embedding task instructions (opens in a new tab)
Embedding controls
Multi-vector compression: blosc2 or zstd, independent of dense encoding_format; omitted compression returns an uncompressed matrix.Source for Embedding controls (opens in a new tab)
Video frame sampling
fps 0.2–5; max_video_tokens 10,240–204,800; min_frame_tokens 16–128; max_frame_tokens 128–640 and ≥minimum; min_frames 5–16.Source for Video frame sampling (opens in a new tab)

Sampling controls do not replace the model context limit. Defaults are model-managed when omitted.

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 15 Oct 2026
CategoryUsageRateUnit
Embedding inputsText input1CN¥0.701M tokens
Embedding inputsImage / video frame input2CN¥1.801M tokens

Rate conditions

  1. Native Doubao-embedding-vision family tariff, applied to this documented endpoint. Input embeddings only; no output-text charge or per-question quote.
  2. Native image-input token tariff. Video metering depends on extracted frames; no flat per-video or audio charge inferred.
Pricing source (opens in a new tab)

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

BytePlus VikingDB vector search

Provider source (opens in a new tab)
API model ID
Skylark-embedding-vision
Access
Existing vector collection/index required. Do not inherit domestic CNY rates or Ark endpoint retirement dates.
Regions
Not collected
API model selector
model_name=Skylark-embedding-vision; model_version=251215Source for API model selector (opens in a new tab)
Input modalities
Text, Image, VideoSource for Input modalities (opens in a new tab)
Embedding dimensions
1024, 2048Source for Embedding dimensions (opens in a new tab)

Default 2048.

Supported input workflows
128k text truncation threshold in the search catalog; mixed text/image/video sequence supported.Source for Supported input workflows (opens in a new tab)
Embedding output modes
Dense, sparse and tensor representations supported in this search interface.Source for Embedding output modes (opens in a new tab)
Embedding task instructions
Instructions supported; automatic query prefixes depend on the selected model version and retrieval modality.Source for Embedding task instructions (opens in a new tab)
API endpoint URL
POST /api/vikingdb/data/search/multi_modalSource for API endpoint URL (opens in a new tab)

Endpoint pricing has not been collected. Check the provider for applicable rates.

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

BytePlus ModelArk · standard

skylark-embedding-vision-251215
Not verified
How it works
Not collected
Minimum reusable input
Not collected
Lifetime & refresh
Not collected
How to enable it
Not collected
Usage fields
Not collected
Billing rules
Not collected
Limits & data boundaries
Not collected

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Volcengine Ark

doubao-embedding-vision-251215
Not verified

No model-specific prompt caching or cached-input tariff established. Vector database storage is a separate product charge.

How it works
Not collected

No model-specific prompt caching or cached-input tariff established. Vector database storage is a separate product charge.

Minimum reusable input
Not collected

No model-specific prompt caching or cached-input tariff established. Vector database storage is a separate product charge.

Lifetime & refresh
Not collected

No model-specific prompt caching or cached-input tariff established. Vector database storage is a separate product charge.

How to enable it
Not collected

No model-specific prompt caching or cached-input tariff established. Vector database storage is a separate product charge.

Usage fields
Not collected

No model-specific prompt caching or cached-input tariff established. Vector database storage is a separate product charge.

Billing rules
Not collected

No model-specific prompt caching or cached-input tariff established. Vector database storage is a separate product charge.

Limits & data boundaries
Not collected

No model-specific prompt caching or cached-input tariff established. Vector database storage is a separate product charge.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

BytePlus VikingDB vector search

Skylark-embedding-vision
Not verified

No model-specific prompt-cache contract or cached-input rate established.

How it works
Not collected

No model-specific prompt-cache contract or cached-input rate established.

Minimum reusable input
Not collected

No model-specific prompt-cache contract or cached-input rate established.

Lifetime & refresh
Not collected

No model-specific prompt-cache contract or cached-input rate established.

How to enable it
Not collected

No model-specific prompt-cache contract or cached-input rate established.

Usage fields
Not collected

No model-specific prompt-cache contract or cached-input rate established.

Billing rules
Not collected

No model-specific prompt-cache contract or cached-input rate established.

Limits & data boundaries
Not collected

No model-specific prompt-cache contract or cached-input rate established.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Inferred from the named native hosted API offering and cross-check of ByteDance and ByteDance-Seed public repositories. No matching downloadable checkpoint was found. This is not an explicit published weight-license declaration.

Classification evidence (opens in a new tab)

Sources reviewed