VoyageAI by MongoDB: voyage-multimodal-3.5
Voyage AI voyage-multimodal-3.5 embeds text, images and video frames into a shared retrieval space. It handles interleaved content such as document screenshots, tables and scenes, with 32,000-token inputs and a default 1024-dimensional representation.
About the model
This description covers the model’s capabilities. Available in Ailoot inputs, outputs and options are listed for the selected method below.
voyage-multimodal-3.5 is Voyage AI’s embedding model for retrieval across text, images and video. It processes interleaved text and visual content through a single transformer encoder, representing a complete multimodal input in a shared vector space. The 3.5 edition adds explicit video-frame support and flexible embedding dimensions.
Preserve the relationship between text and visuals
Prepare a document page or section in the form that preserves its useful visual information. A slide containing a chart and a short explanation can be represented with the slide image and accompanying text in the same ordered input. For a PDF-based collection, the documented visual input is a page screenshot or other image representation of the document.
For a report chart, retain the title, axis labels and legend in the image. Attach explanatory text when it provides the context needed to interpret the visual. Store the page number and source identifier beside the resulting vector so a retrieved candidate can open the relevant page rather than merely return a document name.
The maker’s input format supports sequences of text, images and videos. Its published image formats are PNG, JPEG, WebP and GIF, with MP4 for video. Visual content can be represented through URLs or Base64 data in the documented recipe; keep one representation style consistently within a request.
Retrieve a scene rather than an entire long video
The model represents video as an ordered sequence of frames. The maker recommends dividing a long video into related scenes, aligning scene boundaries with available transcript timestamps and retaining the corresponding transcript text. Lower resolution or a reduced frame rate can help a semantically continuous scene fit its context budget.
For a training-video collection, a query might ask for the scene that demonstrates adjusting a particular control. Associate each indexed scene with its time range. Return the matching scene and inspect the frames or original clip for the actual action, rather than treating the retrieved vector as a written description of everything in the recording.
Transcript text can contribute when prepared as part of the multimodal input. Video-frame embedding itself does not supply a speech transcript.
Context and preparation limits
The stated context length is 32,000 tokens per input. Voyage counts every 560 image pixels and every 1120 video pixels as a token in its published calculation, so visual content shares that budget with text. The documented limits also include 16 million pixels and 20 MB per image, and 20 MB per video. A list contains at most 1000 inputs and 320,000 combined tokens.
The default truncation setting clips an over-length input; if that boundary falls within an image, the entire image is discarded. Preparing complete pages or scenes within the budget therefore preserves more predictable input content. The alternative documented setting rejects over-length inputs instead of truncating them.
Vector dimensions and retrieval roles
The default output dimension is 1024, with 256, 512 and 2048 dimensions also documented for this edition. The announcement describes Matryoshka embeddings and floating-point, signed or unsigned 8-bit, and binary representations. Retain the chosen dimension and representation with the index.
Voyage’s retrieval recipe distinguishes query and document inputs. A query can be text or an image, and indexed documents can combine text and visuals. One vector represents each complete input sequence; store its source page, image or scene association so the ranked result remains inspectable.
Use the returned visual source
A retrieval stage can locate a report page, table, slide or scene relevant to the query. The answer-writing or analysis stage then reads that source material. For a table, inspect the actual headers and values before using a result; for a scene, inspect the relevant frames and associated text.
The substantive difference from voyage-multimodal-3 is the 3.5 edition’s explicit video support and additional representation choices. These are embedding capabilities for locating source material, with numerical vectors as output rather than generated images, video or a narrative answer.
Ailoot Ultra · Embeddings POST /v1/embeddings voyageai/voyage-multimodal-3.5
voyageai/voyage-multimodal-3.5
voyageai/voyage-multimodal-3.5Overview
Voyage AI voyage-multimodal-3.5 embeds text, images and video frames into a shared retrieval space. It handles interleaved content such as document screenshots, tables and scenes, with 32,000-token inputs and a default 1024-dimensional representation.
- Accepted inputs
- Text · token_ids
- Results
- embedding
Available in Ailoot · inputs & methods
Ailoot Ultra · Embeddings POST /v1/embeddings
- Input types
text,token_ids- Output types
embedding- Context
- 32,000 tokens
- Output format
- json
| Field | Type | Requirement | Values & limits | Default |
|---|---|---|---|---|
Fieldmodel | Typestring | RequirementRequired | Values & limits voyageai/voyage-multimodal-3.5 | 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— |
Input material combinations
text_to_embedding
Method parameters and lifecycle
| 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.
Pricing
Tariff references do not add request options. Supported parameters are listed for the selected method.
- Fast mode
- Standard rate When fast mode is selected
- Flex mode
- Standard rate When flex mode is selected
- Maximum reasoning effort
- Standard rate When maximum reasoning effort uses the default adjustment
Ailoot rates
Prices by configuration
| Configuration | Rate schedule | Charge |
|---|---|---|
| ConfigurationBase token rates | Rate scheduleBase rate | Charge Input · Input tokens $0.00024 / 1K tokens Output · Output tokens $0 / 1K tokens Input · Cache read — / 1K tokens Input · Cache write — / 1K tokens Input · Cache write, 1 hour — / 1K tokens |
Ailoot reserves an estimated amount before a request. After the result, it settles the final charge from measured usage. Streaming settlement uses the final usage record. How reservation and settlement work →
Example requests
Ailoot Ultra · Embeddings
set -eu
: "${AILOOT_API_KEY:?Set your Ailoot API key}"
python3 - <<'PY_REQUEST'
import json,os
from pathlib import Path
p=json.loads('{"model": "voyageai/voyage-multimodal-3.5", "input": ["First document", "Second document"]}')
Path("request.json").write_text(json.dumps(p,ensure_ascii=False))
PY_REQUEST
curl --fail-with-body --silent --show-error https://api.ailoot.ai/v1/embeddings \
-H "Authorization: Bearer $AILOOT_API_KEY" -H 'Content-Type: application/json' \
--data-binary @request.json -D response-headers.txt -o result.json
python3 - <<'PY_RESULT'
import json
from pathlib import Path
r=json.loads(Path("result.json").read_text())
for x in r.get("data",[]): print(json.dumps({"index":x.get("index"),"embedding":x.get("embedding")}))
PY_RESULT
from openai import OpenAI
import os
client=OpenAI(base_url="https://api.ailoot.ai/v1",api_key=os.environ["AILOOT_API_KEY"],max_retries=0)
result=client.embeddings.create(model='voyageai/voyage-multimodal-3.5',input=["First document","Second document"])
print(result)
for item in result.data:print(item.index,item.embedding)
Receive and save results
Ailoot Ultra · Embeddings
json
Keep all completed results
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.
If the original response needs checking
No embedding job endpoint is introduced.
Response fields
| Field | Type |
|---|---|
Fieldobject | Typestring |
Fieldmodel | Typestring |
Fielddata[].index | Typeinteger |
Fielddata[].embedding | Typefinite float[] |
Fieldusage | Typeobject |
Errors and next steps
| 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. |
Aliases
Model aliases
Ailoot Ultra voyageai/voyage-multimodal-3.5
Method path aliases
Use the method paths shown above.
Related caller IDs
FAQ · VoyageAI by MongoDB: voyage-multimodal-3.5
- What inputs and results are available?
POST /v1/embeddings·voyageai/voyage-multimodal-3.5:text,token_ids→embedding- Which fields does the request use?
POST /v1/embeddings·voyageai/voyage-multimodal-3.5:model,input,encoding_format,provider- What limits apply?
POST /v1/embeddings·voyageai/voyage-multimodal-3.5:ctx:32000;input.budget.maximum:32000(UTF8 bytes);input.min_items:1