Skip to main content

GLM-5.3-FlashX Understanding

GLM-5.3-FlashX is Zhipu's high-speed serving tier of GLM-5.3-Flash (up to 200 tokens/s, launched 18 Sep 2026) via Z.ai's OpenAI-compatible Chat Completions API. OpenTryOn integrates it via GLMUnderstandAdapter for image and video understanding.

Unlike Zhipu's text-only GLM-5.3 flagship, the Flash/FlashX branch is a vision-language model. Like Kimi and Qwen3.8, it is general-purpose -- useful for describing garments, outfits, and lookbook/runway videos in the fashion domain, but equally capable on documents, UI screenshots, product photography, or any other visual content.

Capabilities​

Capabilityglm-5.3-flashxNotes
ModalitiesText, image, video inText out
Context1M tokens128K max output
ThinkingAlways onthinking.type only supports enabled; use reasoning_effort to control depth instead
Reasoning depthreasoning_effort: low, high, max (default max)Trades thoroughness vs latency/cost
ServingSame weights as GLM-5.3-Flash, faster inference tier~100K domestic accelerators behind the speedup

Prerequisites​

  1. Z.ai account and API key
  2. Set ZAI_API_KEY in your .env file
  3. Optional base URL override via ZAI_BASE_URL
ZAI_API_KEY=your_zai_api_key
# ZAI_BASE_URL=https://api.z.ai/api/paas/v4  # default

Installation​

The GLM adapter reuses the openai package (already a core dependency of opentryon, since the Z.ai API is OpenAI-SDK compatible). No additional installation is required.

Quick Start​

Image Understanding​

from tryon.api.zai import GLMUnderstandAdapter

adapter = GLMUnderstandAdapter()  # glm-5.3-flashx by default

result = adapter.understand_image(
    "garment.jpg",
    prompt="Describe this outfit: color, pattern, style, fit, and material.",
)
print(result["text"])

Video Understanding​

result = adapter.understand_video(
    "runway_clip.mp4",
    prompt="Summarize the styling and garments shown in this video.",
)
print(result["text"])

Public https:// media URLs are passed through; local files are inlined as base64 data URIs.

CLI​

opentryon understand --model glm-5.3-flashx \
  --image garment.jpg --prompt "Describe this outfit."

opentryon understand --model glm-5.3-flashx \
  --video runway_clip.mp4 --prompt "Summarize the styling." \
  --reasoning-effort low

MCP​

Same registry model appears as an MCP tool (no extra wiring):

  • understand_glm_5_3_flashx — Z.ai API (ZAI_API_KEY)

See MCP Server and mcp-server/README.md.

API Reference​

GLMUnderstandAdapter​

class GLMUnderstandAdapter:
    def __init__(
        self,
        api_key: Optional[str] = None,       # ZAI_API_KEY
        model: str = "glm-5.3-flashx",
        base_url: Optional[str] = None,      # ZAI_BASE_URL or the Z.ai default
    )

Methods​

  • understand_image(image, prompt, reasoning_effort=None, max_tokens=None, temperature=None, top_p=None)
  • understand_video(video, prompt, ...)
  • understand(image=None, video=None, prompt=..., ...) — CLI / MCP entry point

Return dict keys: text, reasoning, model, usage.

References​