[Models](/models)/[Captioning](/models?category=captioning)/yanziang/InternVideo3-8B-Instruct

HFScene CaptioningApache 2.0

# InternVideo3-8B-Instruct

by yanziang

Long-video reasoning model with a 256K-token context, built for hour-scale footage and agentic tool use

1.4Kdl/month

9.36B (36-layer language model plus a 27-layer ViT)params

[HuggingFace](https://huggingface.co/yanziang/InternVideo3-8B-Instruct)[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

Identifiers

Model ID

`yanziang/InternVideo3-8B-Instruct`

## Deploy InternVideo3-8B-Instruct

Single-tenant

Mixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.

[Read the customer-hosted guide](https://mixpeek.com/docs/customer-hosted/overview) [Talk to us about single-tenant](/demo)

## Overview

InternVideo3-8B-Instruct reads video and writes text. It is built around Multimodal Contextual Reasoning, which keeps perception, planning, tool calls, self-checking and memory inside one shared context so the model can take several passes over a long clip instead of a single forward look.

The architecture choice that makes hour-scale footage tractable is M²LA, a multi-head latent attention that compresses the KV cache to a rank-896 latent per layer. That is what buys a 262,144-token context without dropping frames on the way in. Training ran a short-to-long curriculum, ending at 4fps over 2,048 frames, then rule-based reinforcement learning on temporal grounding and video QA, then on-policy distillation from Qwen3-235B.

One practical note before you copy the quickstart: the weights live under the first author's personal namespace at yanziang/InternVideo3-8B-Instruct, while the card's own snippet points at OpenGVLab/InternVideo3-8B-Instruct, which does not publicly resolve. Swap the path.

## Architecture

InternVideo3ForConditionalGeneration, 9,364,294,384 parameters in BF16. A 27-layer ViT at hidden size 1,152 with patch size 16 and temporal patch size 2 feeds a 36-layer language model at hidden size 4,096 with 32 attention heads. M²LA holds a KV latent rank of 896 per layer, and the maximum context is 262,144 tokens. Loaded through Transformers with trust\_remote\_code, since the modeling code ships in the repo rather than upstream.

## Mixpeek SDK Integration

IngestRetrieve

// InternVideo3 writes text, so on Mixpeek it belongs at ingest: describe the
// video with the model, then store the description and its embedding.
// Route verified against the live OpenAPI:
// POST /v1/namespaces/{namespace\_id}/documents/upsert
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns\_your\_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API\_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection\_id: "col\_your\_collection",
      documents: \[
        {
          document\_id: "match-0091-seg-14",
          vectors: { "text-embedding": descriptionVector },
          payload: {
            description: internVideo3Output,
            start\_s: 842,
            end\_s: 871,
          },
          metadata: { source: "match-archive" },
        },
      \],
    }),
  },
);

// For a managed path that needs no self-hosting, universal\_extractor@v1 runs
// description, transcription and embedding over a bucket in one pass.

## Capabilities

*   262,144-token context over video, images and text
*   Trained to 2,048 frames at 4fps through a short-to-long curriculum
*   Temporal grounding, so an answer can carry the time range it came from
*   Tool use across temporal grounding, ASR, web search and video segmentation
*   English and Chinese
*   Apache 2.0, with no usage restriction on self-hosting

## Use Cases on Mixpeek

Describing long-form footage at ingest so the text becomes searchable

Answering a question about a full episode or match without pre-chunking it

Locating the moment an event happens and returning its start and end

Reading the candidates a cheaper embedding model retrieved, then answering with citations

## Performance

Input SizeVideo to 2,048 frames, images, or text; 262,144-token context

GPU LatencyInput dependent

GPU ThroughputBatch dependent

GPU Memory~19 GB in BF16 for weights alone

No latency or throughput figures are published for this model. The memory number is computed from the 9,364,294,384 BF16 parameters in the safetensors index and covers weights only. A full 256K-token context needs substantial KV cache on top, even with M²LA's rank-896 factorization.

## Common Pipeline Companions

[

tencent/WeMM-Embedding-9B

First-stage multimodal retrieval before this model reads the candidates

](/model/tencent/WeMM-Embedding-9B)[

OpenGVLab/InternVideo2-Stage1-1B-224p-K700

Embedding model from the same lineage for the retrieval half

](/model/OpenGVLab/InternVideo2-Stage1-1B-224p-K700)[

Qwen/Qwen3-VL-8B-Instruct

Alternative video-capable VLM with published benchmark numbers

](/model/Qwen/Qwen3-VL-8B-Instruct)

## Frequently Asked Questions

### Why does the model card's quickstart fail?

It sets model\_path to OpenGVLab/InternVideo3-8B-Instruct, and that repository does not publicly resolve: the Hugging Face API returns 401 for it, and a hub-wide search for InternVideo3 returns only yanziang/InternVideo3-8B-Instruct. Checked 2026-08-27. Point the path at the yanziang repo and the rest of the snippet works unchanged.

### How long a video can InternVideo3-8B-Instruct actually handle?

The frame budget binds before the token count does. Training tops out at 2,048 frames, so 4fps gives you about 8.5 minutes of footage and an hour of material means sampling at roughly 0.5fps. The card's hour-scale claim and its 4fps training figure are both true, at different sampling rates. Pick the rate from how fast the thing you are looking for happens.

### Can InternVideo3 power search on its own?

No. It has no embedding head, so it cannot populate a vector index. Its place in a retrieval system is at ingest, turning footage into descriptions and grounded time ranges that a text or multimodal embedding model then indexes, or as the second stage that reads whatever first-stage retrieval returned.

### Are there published benchmark scores?

The model card carries no benchmark table, only architecture and training details. The numbers are in the arXiv paper, 2606.12195. That absence is worth knowing before you plan an evaluation: there is no card-level score to compare against, so budget for scoring it on your own footage.

### Can I run it on Mixpeek today?

Not as a managed extractor. GET /v1/discovery/extractors returns 13 shipping extractors and none accepts an arbitrary Hugging Face model id. Run it yourself and upsert descriptions and vectors through POST /v1/namespaces/{namespace\_id}/documents/upsert, or use universal\_extractor@v1 for a managed describe-and-embed pass that needs no GPUs of your own.

## Explore on Mixpeek

[

More Captioning models

Compare alternatives in this category

](/models?category=captioning)[

Best Document AI Platforms

Hand-picked tools & platforms compared

](/curated-lists/best-document-ai-platforms)[

OCR & Document AI Internals

Deep-dive technical guide

](/guides/ocr-document-ai-internals)[

Feature Extractors

See how Mixpeek runs models as extractors

](/extractors)[

Mixpeek Vector Store

Store & search embeddings at scale

](/mvs)[

Pricing

Usage-based pricing for pipelines

](/pricing)[

All Curated Lists

Compare models, APIs & infrastructure

](/curated-lists)

### Specification

FrameworkHF

Organizationyanziang

FeatureScene Captioning

Outputtext

Modalitiesvideo, image

RetrieverSemantic Search

Parameters9.36B (36-layer language model plus a 27-layer ViT)

LicenseApache 2.0

Downloads/mo1.4K

[

### Research Paper

InternVideo3: Agentify Foundation Models with Multimodal Contextual Reasoning

arxiv.org](https://arxiv.org/abs/2606.12195)

### Build a pipeline with InternVideo3-8B-Instruct

Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

### Alternative Models

[

Salesforce/blip2-opt-2.7b

Scene Captioning

](/model/Salesforce/blip2-opt-2.7b)[

microsoft/Florence-2-large

Scene Captioning

](/model/microsoft/Florence-2-large)[

google/paligemma2-3b-mix-448

Scene Captioning

](/model/google/paligemma2-3b-mix-448)[

Qwen/Qwen3-VL-8B-Instruct

Scene Captioning

](/model/Qwen/Qwen3-VL-8B-Instruct)