zai-org / zai-org/feedback

[建议 / Feature] 多模态模型(GLM-5.3-Flash)会话下 Read 图片仍强制转 CDN URL + 4.5V 委托,期望图片直进上下文

Open
#615 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

priority: P2
Dominant language
No language data
Stars
22
Forks
1
PR merge metrics
No merged PRs in 30d

Description

现状

GLM-5.3-Flash 在模型注册中声明输入模态为 text + image + video(多模态)。但实测把 ZCode 会话模型切换到 GLM-5.3-Flash 后,Read 工具读图片依然被上传 CDN 返回 URL,再由内置 analyze_image(GLM-4.5V)委托看图——图片像素始终不进入主模型上下文。

问题

Flash 的原生视觉能力(官方定位:主动观察界面 / 渲染 / 交互反馈的视觉编码闭环)在 ZCode 内实际用不起来,多模态模态声明形同虚设;每次看图多一跳委托,慢且受 4.5V 质量上限约束。

期望

当会话模型的 modalities 包含 image 时,Read / 拖拽 / 粘贴的图片直接作为视觉输入进入主模型上下文(对标 Claude Code / Codex 的原生多模态体验),委托 MCP 仅作为纯文本模型的 fallback。

环境

ZCode 桌面客户端(macOS, Apple Silicon),GLM coding plan,分别实测 GLM-5.3 与 GLM-5.3-Flash 会话。

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The issue names no files, tests, or implementation entry points. Start by tracing how Read, drag-and-drop, and pasted images are handled when the session model's modalities include image, including the analyze_image delegation path. Done means images reach the main multimodal model context directly, while delegation remains a fallback for text-only models.

Written by the indexing model from the issue text.

Assessment

Domain
ai, desktop
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.