Glint-VL Multimedia Understanding API

Overview

Glint-VL-Video-8B is an enterprise-grade vision-language model that provides unified understanding of images, videos, and spatial information. The API uses an OpenAI-compatible format, allowing you to analyze remote images or videos with natural-language prompts and receive text results ready for use in business workflows.

Quickstart

First, list the available models:

curl https://maas.ve2s.ai/v1/models \
  -H "Authorization: Bearer $GLINT_API_KEY"

Analyze a video with Glint-VL-Video-8B:

curl -X POST https://maas.ve2s.ai/v1/chat/completions \
  -H "Authorization: Bearer $GLINT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Glint-VL-Video-8B",
    "messages": [{
      "role": "user",
      "content": [
        {
          "type": "video_url",
          "video_url": {
            "url": "https://example.com/video.mp4"
          }
        },
        {
          "type": "text",
          "text": "Summarize the main events and character actions in the video"
        }
      ]
    }],
    "stream": false
  }'

Authentication

Both endpoints use a Bearer API key following the OpenAI standard. Add the key assigned by your service provider to the Authorization request header:

Authorization: Bearer <GLINT_API_KEY>
Content-Type: application/json

The examples use the GLINT_API_KEY environment variable. Do not commit your key to a public repository or embed it in a frontend page.

Base URL

https://maas.ve2s.ai/v1

This document uses https://maas.ve2s.ai/v1 as an example. Replace it with the API base URL assigned by your service provider when making requests.

Models Endpoint

Returns the models supported by the media analysis endpoint.

Request

No request body is required, but a Bearer API key must be provided:

curl https://maas.ve2s.ai/v1/models \
  -H "Authorization: Bearer $GLINT_API_KEY"

Successful Response

The response follows the OpenAI Models API format:

{
  "object": "list",
  "data": [
    {
      "id": "Glint-VL-Video-8B",
      "object": "model",
      "created": 0,
      "owned_by": "glint"
    },
    {
      "id": "Glint-VL-SE-8B",
      "object": "model",
      "created": 0,
      "owned_by": "glint"
    }
  ]
}

The returned data[].id can be used directly as the model parameter in a POST /v1/chat/completions request.

Media Analysis Endpoint

Analyzes images or videos and supports follow-up questions.

Purpose

The request and response bodies are compatible with the OpenAI Chat Completions format. The initial request analyzes the specified media. For follow-up requests, you can reuse the server-side session through conversation_id, or maintain stateless conversation history by sending the complete messages array on each request.

Request Parameters

Content-Type: application/json

Request Examples

Analyze an Image

Glint-VL-SE-8B supports images only. Glint-VL-Video-8B can use the same image_url structure:

curl -X POST https://maas.ve2s.ai/v1/chat/completions \
  -H "Authorization: Bearer $GLINT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Glint-VL-SE-8B",
    "messages": [{
      "role": "user",
      "content": [
        {
          "type": "image_url",
          "image_url": {
            "url": "https://example.com/image.jpg"
          }
        },
        {
          "type": "text",
          "text": "Describe the image in detail and extract all text"
        }
      ]
    }],
    "stream": false
  }'

Ask a Follow-up Question

Use the conversation_id from the initial response to ask a follow-up question without sending the media again. The model should match the model used by the session:

curl -X POST https://maas.ve2s.ai/v1/chat/completions \
  -H "Authorization: Bearer $GLINT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Glint-VL-Video-8B",
    "conversation_id": "conv_xxxxxxxx",
    "messages": [{
      "role": "user",
      "content": "Where did the person wearing red go afterward?"
    }],
    "stream": false
  }'

For a fully stateless workflow consistent with OpenAI Chat Completions, omit conversation_id and resend the initial user message, the initial assistant message, and the latest user message in order within messages.

Streaming Request

For streaming requests, use curl -N to disable client-side buffering:

curl -N -X POST https://maas.ve2s.ai/v1/chat/completions \
  -H "Authorization: Bearer $GLINT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Glint-VL-Video-8B",
    "messages": [{
      "role": "user",
      "content": [{
        "type": "video_url",
        "video_url": {
          "url": "https://example.com/video.mp4"
        }
      }]
    }],
    "stream": true
  }'

Successful Response

When stream=false, the endpoint returns a complete Chat Completion:

{
  "id": "chatcmpl-xxxxxxxx",
  "object": "chat.completion",
  "created": 1784736000,
  "model": "Glint-VL-Video-8B",
  "conversation_id": "conv_xxxxxxxx",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Media analysis result",
        "refusal": null,
        "annotations": []
      },
      "logprobs": null,
      "finish_reason": "stop"
    }
  ]
}

Field Descriptions

SSE Response

When stream=true, the response type is text/event-stream. Content is returned incrementally through choices[0].delta.content, followed by a final data: [DONE] event:

data: {"id":"chatcmpl-xxxxxxxx","object":"chat.completion.chunk","created":1784736000,"model":"Glint-VL-Video-8B","choices":[{"index":0,"delta":{"role":"assistant","content":""},"logprobs":null,"finish_reason":null}]}

data: {"id":"chatcmpl-xxxxxxxx","object":"chat.completion.chunk","created":1784736000,"model":"Glint-VL-Video-8B","choices":[{"index":0,"delta":{"content":"Media analysis result…"},"logprobs":null,"finish_reason":null}]}

data: {"id":"chatcmpl-xxxxxxxx","object":"chat.completion.chunk","created":1784736000,"model":"Glint-VL-Video-8B","choices":[{"index":0,"delta":{},"logprobs":null,"finish_reason":"stop"}]}

data: [DONE]

Error Response

The error body uses the OpenAI-style format:

{
  "error": {
    "message": "Error description",
    "type": "invalid_request_error",
    "param": "model",
    "code": "invalid_model"
  }
}

Notes

• The endpoint does not accept multipart file uploads or local file paths. Use image_url for images and video_url for videos.
• Images support HTTP/HTTPS URLs and Base64 Data URLs. Videos support HTTP/HTTPS URLs only.
• The entire JSON request body is limited to 10 MB. Remote image downloads for Glint-VL-SE-8B are limited to 20 MB.
• Glint-VL-SE-8B accepts image_url only. Glint-VL-Video-8B accepts either image_url or video_url.
• messages[].role supports developer, system, user, and assistant. Media content blocks may appear only in user messages.
• Each user message may contain at most one image_url or video_url. Combined text content is limited to 100,000 characters.
• When uploading new media with a conversation_id, place the new media and the current question in the final user message.
• Non-streaming content is returned in choices[0].message.content. Streaming content is returned in choices[0].delta.content.
• The server waits for up to 12 minutes. During analysis, streaming requests receive an SSE heartbeat comment every 15 seconds.
• Streaming deployments must disable reverse-proxy response buffering. If the client disconnects before completion, the background analysis is not canceled.

SDK Examples

All examples below use Glint-VL-Video-8B to analyze a video. First, set the API key in the GLINT_API_KEY environment variable:

import os

import requests

payload = {
    "model": "Glint-VL-Video-8B",
    "messages": [{
        "role": "user",
        "content": [
            {
                "type": "video_url",
                "video_url": {"url": "https://example.com/video.mp4"},
            },
            {
                "type": "text",
                "text": "Summarize the main events and character actions in the video",
            },
        ],
    }],
    "stream": False,
}

response = requests.post(
    "https://maas.ve2s.ai/v1/chat/completions",
    headers={"Authorization": f"Bearer {os.environ['GLINT_API_KEY']}"},
    json=payload,
    timeout=720,
)
response.raise_for_status()

data = response.json()
print(data["choices"][0]["message"]["content"])
print("conversation_id:", data["conversation_id"])

Frequently Asked Questions

© 2026 VE²S · Glint-VL · Enterprise Vision Intelligence Model