NEWVectors or files. Pick a path.Start →

    Models

    Every model Mixpeek runs, calls or can load: what it is good and bad at, its inputs and outputs, how it runs with scale to zero, its license, and the templates that use it. Ordered by use. Filter by anything on the record; the address bar keeps the filter so you can share it.

    91 models

    Templates models.json

    Recommended by use case

    Task: Video and image Q&A

    6 of 91 models

    Hardware, cost and evidence come from each record. "Unmeasured" means we have not measured it on our own image yet.

    • Provider API

      Gemini 3.6 Flash

      Video and image Q&A · gemini-3.6-flash

      High-quality video QA and structured tagging.

      • In/out Video, Audio, Image, Text, Document to Text, Structured data
      License
      Proprietary
      Hardware
      Provider API
      View modelJSON
    • Provider API

      Gemini 3.1 Flash-Lite

      Video and image Q&A · gemini-3.1-flash-lite

      Low-cost bulk tagging of scenes and frames.

      • In/out Video, Audio, Image, Text, Document to Text, Structured data
      License
      Proprietary
      Hardware
      Provider API
      View modelJSON
    • Provider API

      Gemini 2.5 Flash-Lite

      Video and image Q&A · gemini-2.5-flash-lite

      Cheapest Gemini video tagging and captioning at 1 fps.

      • In/out Video, Audio, Image, Text, Document to Text, Structured data
      License
      Proprietary
      Hardware
      Provider API
      View modelJSON
    • Provider API

      Gemini 3 Flash (preview)

      Video and image Q&A · gemini-3-flash-preview

      Stronger video reasoning than 2.5 Flash at Flash latency.

      • In/out Video, Audio, Image, Text, Document to Text, Structured data
      License
      Proprietary
      Hardware
      Provider API
      View modelJSON
    • Provider API

      Gemini 3.8 Flash

      Video and image Q&A · gemini-3.8-flash

      Brand, artist, venue and moment tags per shot (the default tag model in a custom shot plugin).

      • In/out Video, Audio, Image, Text, Document to Text, Structured data
      License
      Proprietary
      Hardware
      Provider API
      View modelJSON
    • Not runnable today

      InternVideo2.5 Chat 8B

      Video and image Q&A · OpenGVLab/InternVideo2_5_Chat_8B

      Long-video question answering and fine-grained temporal grounding with an open model (long and rich context modeling).

      • In/out Video, Text to Text
      License
      Apache-2.0
      Hardware
      Needs a GPU
      View modelJSON

    Questions

    Which of these models does Mixpeek run today?
    15 are served on Mixpeek as Ray Serve apps that start at zero replicas and wake on first use (Multilingual E5 Large Instruct, all-MiniLM-L6-v2, SigLIP Base patch16 224, InsightFace buffalo_l (SCRFD-10GF + ArcFace ResNet50 WebFace600K), InsightFace buffalo_sc (SCRFD-500MF + MobileFaceNet WebFace600K) and others). 13 are provider APIs Mixpeek calls by name. 34 can run as a custom plugin and do not run yet. 29 are blocked today; each page says why.
    How is the default order decided?
    By usage, the namespaces and collections that use each model, counted on 2026-11-06T13:42:25Z, then by how many templates use it, whether it runs today and whether we have our own evaluation.
    What does unmeasured mean on a model?
    We have not measured that model's cost or quality on our own image and data yet. Publisher benchmarks are linked from the model's sources and are not copied into the measured fields.
    Can an agent read this catalog?
    Yes. /models.json holds every record and /models/<id>.json holds one. They are the same records this page renders, generated from the catalog the Mixpeek server is built from.
    How do I use a model in a namespace?
    Open the model and deploy a template that already uses it, or add it to a new or existing namespace from the Designer in Studio. Every served model starts at zero replicas, so an idle namespace costs nothing for inference.

    Hugging Face model directory

    Directory pages for 422 public models. They are not in the catalog above, and Mixpeek does not run them as listed.