Best Chat UI for GLM: Open WebUI and Cherry Studio
The best chat UI for GLM depends on who uses it. Open WebUI setup steps, where Cherry Studio takes a base URL, which GLM ids to add, and fixes.
By OceanToken Team

Editor's note: Most GLM complaints in chat apps are not about the app. They come from forced thinking, a text-only model getting a pasted image, or a model list that never loads. Our pick: Open WebUI for a team, Cherry Studio for one person. In either one, set up two GLM entries, not one. We have not run each UI ourselves; the steps below come from each tool's docs.
Short answer: The best chat UI for GLM is Open WebUI if you run one shared install for a team, and Cherry Studio if you are one person on a desktop. Both need the same three things: a base URL, an API key and a model id. Add a fast Flash entry as the default and the flagship as an opt-in.
How to pick the best chat UI for GLM
Every tool here accepts an OpenAI-compatible URL, so GLM runs in all of them. The real differences are who hosts the app and who manages the keys.
| If you need | Pick | Check before you commit |
|---|---|---|
| One shared web UI for a team | Open WebUI | Admin connections, which users see which models |
| A desktop app for one person | Cherry Studio | Where keys are stored, manual model ids |
Four checks apply to every tool:
- Can you type a model id by hand when the list is empty?
- Can you add custom request fields, such as
reasoning_effort? - Does it show or hide reasoning text?
- Can you mark which models accept images?
A tool that fails the first two is awkward for GLM.
Open WebUI with GLM, step by step
These steps follow Open WebUI's guide for OpenAI-compatible providers:
- Go to Settings > Admin > Connections and find Manage OpenAI API Connections.
- Click Add Connection.
- Set the URL to
https://api.oceantoken.ai/v1and paste your key. - If models do not appear, add ids under Model IDs: type the id, click +, then save.
- Click Verify Connection, then Save.
Two details trip people up. First, the docs say saving a connection does not test it, so use Verify. Second, Verify can fail with 400, 401 or 403 on providers without a standard /models endpoint, and chat may still work. If that happens, add the ids by hand and send a test message.
The docs also note a Docker detail. Replace localhost with host.docker.internal only when the model server runs on your own machine. A hosted API URL does not need it.
Cherry Studio setup and other front ends
Cherry Studio. Open Settings, then the Model Services tab, and click + Add Provider. Enter a name, your key and the API address. Then click Get model list, or use + to add a model id by hand. The docs do not say whether the app appends /v1 for you. If the first message returns a 404, try the URL with and without /v1.
Other front ends. Other chat UIs, such as LobeChat, may accept a custom OpenAI-compatible provider too. Menu names change between versions, so follow the docs for the version you run.
Which GLM ids to add
These ids come from our catalog on 2026-11-02.
| Id | Use in chat | Input |
|---|---|---|
z-ai/glm-5.3-flash |
Everyday chat, pasted screenshots | Text, image, video |
z-ai/glm-5.3 |
Hard questions, long code | Text only |
z-ai/glm-4.7 |
Quick replies with thinking off | Text only |
The flagship is listed as text only on the GLM-5.3 model page. If someone pastes an image into a chat bound to it, expect an error or an ignored image. Bind image work to the Flash entry.
Why GLM pauses before it answers
Z.ai's docs say GLM-5.3 and GLM-5.3-Flash use forced thinking, and it cannot be disabled. In a chat UI that shows up as a pause before the first word. The GLM-4.7 series can turn thinking off with "thinking": {"type": "disabled"}.
Reasoning comes back as reasoning_content. Some front ends fold it away; others print it in the reply. Check how yours treats it before you roll out.
Our catalog lists reasoning_effort as the control for GLM-5.3. If your UI has a custom parameters box, try reasoning_effort: low on the flagship entry. Whether every front end forwards that field is something to test, not assume.
Common errors and fixes
| Symptom | Likely cause | Fix |
|---|---|---|
| Connection saved, model list empty | No /models call, or a slow list |
Add ids by hand |
| Verify fails, chat works | Non-standard /models |
Ignore Verify, test a message |
| Long pause before reply | Forced thinking | Use Flash, or lower effort |
| Image paste errors | Text-only flagship | Switch to the Flash entry |
| Costs climb in long chats | Full history resent each turn | Start a new chat per topic |
Our take: Make the Flash entry the default for everyone and keep the flagship as an opt-in for a smaller group. Most chat questions do not need the flagship, and switching is one dropdown away. If many chats escalate, flip the default for that group.
How to test this on your workload
- Add the connection in a clean Open WebUI install.
- Check that the model list loads, or add the ids by hand.
- Paste one screenshot into the Flash entry and one into the flagship. Note what happens.
- Ask one hard question on each entry. Note the wait and whether reasoning text shows.
- Run the curl checks below first if anything fails, to rule out the key.
curl https://api.oceantoken.ai/v1/models \
-H "Authorization: Bearer $OCEANTOKEN_API_KEY" | head -c 400
curl https://api.oceantoken.ai/v1/chat/completions \
-H "Authorization: Bearer $OCEANTOKEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "z-ai/glm-5.3-flash", "max_tokens": 50, "messages": [{"role": "user", "content": "hi"}]}'
Running it on OceanToken
Point any of these front ends at https://api.oceantoken.ai/v1 and use z-ai/glm-5.3 or z-ai/glm-5.3-flash as the model id. The reason to use OceanToken in a chat UI is the dropdown: one connection gives your team GLM and other model families on one key and one balance. See the current rate on /glm-api.
FAQ
Why do no GLM models show up? The front end could not read a model list. Add the ids by hand under Model IDs or the + button.
Can I paste images into GLM? Yes, on z-ai/glm-5.3-flash. The flagship z-ai/glm-5.3 is text only.
Can I turn off thinking on GLM-5.3? No. Z.ai says it is forced on GLM-5.3 and GLM-5.3-Flash. Use the GLM-4.7 series if you need it off.
Which is the best chat UI for GLM? Open WebUI for a shared team install, Cherry Studio for one person on a desktop.
Do I need host.docker.internal? Only when Open WebUI runs in Docker and the model server runs on the same host.
Sources
- Open WebUI: Starting with OpenAI-compatible servers (checked 2026-11-02)
- Cherry Studio: Custom provider (checked 2026-11-02)
- Z.ai: Thinking mode (checked 2026-11-02)
- OceanToken GLM-5.3 model page (checked 2026-11-02)
- OceanToken GLM API page (checked 2026-11-02)
Get your OceanToken API key
- Create an OceanToken account and an API key.
- Copy the snippet below, paste your key, and run it.
- Check the current rate on /glm-api.
curl https://api.oceantoken.ai/v1/chat/completions \
-H "Authorization: Bearer $OCEANTOKEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "z-ai/glm-5.3-flash", "messages": [{"role": "user", "content": "Hello"}]}'
More from the blog
A Unified API for Chat, Image and Video
Building on a unified API for chat, image and video: sync vs async calls, billing units, retries, timeouts and storage, with three calls and a launch checklist.
Kimi K3 API International Access Guide
Use the Kimi K3 API internationally: sign up with email, pay by card, create one key and call moonshotai/kimi-k3 with an OpenAI SDK. Steps, code and checks.
Call Tencent Hy3 With the OpenAI SDK
Call Tencent Hy3 with the OpenAI SDK outside China: base URL, model id, streaming, tool calls, reasoning_effort values, Hy3 vs Hy3 preview, and token billing.