NEWVectors or files. Pick a path.Start →
    Provider APIEmbeddingVerified 2026-11-05

    Gemini Embedding 2

    gemini-embedding-2

    Gemini Embedding 2 is a embedding model. Called on Mixpeek through the provider API.

    Mixpeek calls the provider's API by name. Provider terms apply.

    Use in a new namespaceAdd to an existing namespace Record as JSON

    New opens the Designer in Studio. For an existing namespace, open it in Studio and use its Designer tab.

    No demo for Gemini Embedding 2 yet

    A video, image, slideshow or playground goes here once we record one on our own data.

    What it is good and bad at

    Best at
    One vector space for video segments, images, PDFs, audio and text queries; the default for object-level multimodal extractors (multimodal_extractor v2, universal_extractor).
    Worst at
    Cost and latency per call (provider API, 1.9 to 7 s query spikes after a lull per the routes comment); no task_type parameter, so Mixpeek prepends instruction text instead; 3072-d vectors are large to store.
    Task
    Embedding
    Input
    Text, Image, Video, Audio, Document
    Output
    Vector
    Use cases
    Visual search, Text search, Multilingual search, Video understanding, Audio search, Document search

    Templates that use it

    • RTSP Live Monitoring Wall

      Turn always-on RTSP cameras into a searchable monitoring wall where quiet footage is skipped before extraction, kept segments carry gate verdicts, and email alerts fire from retriever results.

      Deploy this template
    • Video Moderation Extraction Foundation

      Segment a video corpus, extract a keyframe record for every segment, and match brand marks against a reference set you upload. Face and public-figure matching run on a dedicated or customer-hosted deployment. This deploy is the extraction layer a moderation policy sits on, and the page carries a state on every part so you can see which policy pieces are yours to add.

      Deploy this template

    Input and output

    Input

    text
    max_tokens
    8192
    image
    inline or URI parts
    video
    segments up to 120 s (multimodal_extractor caps Gemini segments at 120 s)
    prompts
    embedding_task maps to a text prefix: RETRIEVAL_QUERY 'Represent this query for retrieving relevant documents: ', RETRIEVAL_DOCUMENT 'Represent this document for retrieval: ', SEMANTIC_SIMILARITY ...

    Output

    vector
    dims
    3072
    distance
    cosine
    matryoshka
    outputDimensionality when dimensions != 3072

    How Mixpeek wires it

    Extractors
    • multimodal_extractor@v2 fixed
    • universal_extractor@v1 fixed (GEMINI_EMBEDDING_MODEL env, default gemini-embedding-2)
    Feature URIs
    • mixpeek://multimodal_extractor@v2/gemini-embedding-2
    • mixpeek://universal_extractor@v1/gemini-embedding-2
    Vector indexes
    multimodal_extractor_v2_multimodal_embeddinguniversal_extractor_v1_embedding
    Retriever stages
    feature_search
    Combines with
    • multimodal_extractor v2: video segment + Gemini description -> one Gemini Embedding 2 vector, beside E5 transcription/OCR/description vectors

    How it runs

    Status
    Provider API Provider API (Gemini developer API or Vertex). The google__gemini_embedding_2 registry app proxies query embeds; on the planes it is hosted inside mixpeek__api_inference. Extractors call the API directly from the batch step.
    Serve apps
    google__gemini_embedding_2
    Runtime
    Gemini REST embedContent (gemini_rest_endpoint) for queries; google-genai client.models.embed_content in extractors
    Weights
    Provider API, no remote code
    Parameters
    Not recorded
    Size on disk
    Not recorded
    Preprocessing
    Build content parts (text, image, video, PDF), prepend the task instruction to text.
    Postprocessing
    Vector as returned (3072 dims unless dimensions is set); responses cached by model, task, dims and content.
    Serving
    both
    Hardware
    Provider API

    Scale to zero

    Minimum replicas
    0
    Maximum replicas
    6
    Idle timeout
    300 s
    Memory
    Not recorded
    CPUs
    0.5
    GPUs
    0
    Cold start, node warm
    Not recorded
    Cold start, new node
    Not recorded

    Measured cost

    Unmeasured

    No committed throughput+latency+cost measurement; billed per token by Google.

    Our evaluations

    Unmeasured

    We have not evaluated this model on our own data yet. Publisher results are in the sources below.

    Questions

    What is Gemini Embedding 2 best at?
    One vector space for video segments, images, PDFs, audio and text queries; the default for object-level multimodal extractors (multimodal_extractor v2, universal_extractor).
    What are the limits of Gemini Embedding 2?
    Cost and latency per call (provider API, 1.9 to 7 s query spikes after a lull per the routes comment); no task_type parameter, so Mixpeek prepends instruction text instead; 3072-d vectors are large to store.
    Can Gemini Embedding 2 be used commercially?
    Yes. Paid Google API under Google's terms; Mixpeek resells the call and pays per token.
    How does Gemini Embedding 2 run on Mixpeek?
    Provider API (Gemini developer API or Vertex). The google__gemini_embedding_2 registry app proxies query embeds; on the planes it is hosted inside mixpeek__api_inference. Extractors call the API directly from the batch step.
    What does Gemini Embedding 2 output?
    A 3072-dimensional dense vector, compared by cosine similarity.