Vega VL Integration Docs Updated โ
The API documentation for Vega VL, our multimodal understanding model, has been comprehensively expanded. The endpoint and model IDs are unchanged, so existing integrations keep working without modification.
| Model ID | Positioning |
|---|---|
wand-vega-vl-1.0-lite | Cost first โ short-video understanding, high-volume image classification |
wand-vega-vl-1.0-flash | Balanced โ data labeling, summarizing mid-length videos |
wand-vega-vl-1.0-pro | Quality first โ long-video understanding, complex understanding tasks |
1. What's New โ
| Section | Description |
|---|---|
| Choosing a version | Positioning and best-fit scenarios for lite / flash / pro |
| Setup | Three steps: enable the service โ create an API token (console or CreateAigcApiToken) โ call the model |
| Per-key settings | Name, model allowlist, IP allowlist, and quota limit |
| Quick start | Minimal calls with the OpenAI SDK and cURL (streaming) |
| Response fields and usage | Response field reference and token billing details |
| Error codes, usage lookup, token management | HTTP status and business error codes, console / API usage lookup, and token create / delete / query / update APIs |
2. Request Spec Additions โ
- Auth headers: use either
Authorization: Bearer {TOKEN}orx-api-key: {TOKEN}, plus the newX-Request-Idfor request tracing. - Request parameters: added
reasoning_effort(none~xhigh),max_completion_tokens,tools,tool_choice, andjson_schemaforresponse_format. - Media input: audio can be sent as
input_audio.data(Base64) +format, and files asfile.file_data(Base64) +file.file_name.
3. Corrections โ
| Item | Before | Now |
|---|---|---|
stream | Fixed to false | Both true and false supported |
| File (PDF, etc.) input | Pro only | Supported on all three tiers |
| Per-request limit | 10 images and videos combined | Up to 10 videos |
Sending a video as
type: "file"returns502+502003. Always send videos as a video Part.
4. Before You Start โ
A new token takes about
1minute to sync to the gateway; calls made immediately after creation may return401001.Tokens are not interchangeable between the China and international sites, and each application can hold up to
50.Prodoes not accept text-only requests โ useFlashorLitefor your first minimal call.