Configure ZCode Plans and Models
Complete the two basic choices first: built-in plan → everyday model. Add a third-party API only when you need another model. A third-party model is optional, not required.
The Beginner-Friendly Answer
Default: built-in BigModel plan + GLM-5.3-Flash
Need a specific model: add a third-party API afterward
- To start quickly, use a built-in BigModel plan;
- For an initial paid test, you can choose GLM Coding Lite (the screenshot shows plans starting at ¥118);
- For most daily work, start with
GLM-5.3-Flash; - Continue to the third-party section only when you explicitly need an external model.
Required Setup
Complete the next two items and Step 3 is finished.
1. Choose a Built-in Plan
Open the account menu at the lower left and select Upgrade.
For an initial test, start with a built-in BigModel plan instead of buying and configuring several third-party APIs. If you are ready for a paid trial, you can start with GLM Coding Lite.
The screenshot was captured on September 4, 2026 and shows a starting price of ¥118. Pricing and benefits may change, so follow the current page shown in ZCode.
2. Choose a Daily Model
For beginners, this guide recommends GLM-5.3-Flash as the everyday default. It is usually enough for organizing material, reading pages, analyzing screenshots, and making light edits. Switch to a stronger model for long-document synthesis, complex multi-file changes, or when two attempts still produce poor results.
Open the model menu below the prompt box and select:
BigModel → GLM-5.3-Flash
Step 3 is now complete, and you can start a real task. Continue below only when you specifically need another model through a third-party API.
Third-Party Models (Optional)
1. Decide Whether You Need One
Continue only when at least one of these applies:
- You already have an API key from another provider;
- A task specifically requires a particular model;
- You need a capability or model not currently included in the built-in plan.
If none applies, stay with GLM-5.3-Flash.
The next section uses DeepSeek to demonstrate the complete setup. ZCode can also connect to other compatible providers. The workflow is similar, but the URL, API format, and model ID must come from that provider's current documentation.
2. Open Model Settings
Go to:
Gear icon → Settings → Model Settings
Select the existing DeepSeek provider, or click Add Provider to create one.
3. Enter the DeepSeek API Details
Follow the provider's current documentation. This guide uses the following example:
| Setting | Example |
|---|---|
| Provider | DeepSeek |
| Base URL | https://api.deepseek.com |
| API format | Responses (/responses) |
| API Key | Your DeepSeek API Key |
| Model ID | deepseek-v4-flash-vision-exp |
See the DeepSeek Responses API documentation for the interface format. The model ID and Base URL must come from the same provider. If your provider does not list deepseek-v4-flash-vision-exp, do not enter it manually; use a model ID that the provider actually offers.
Treat an API key like a password: it only needs to go into ZCode, so never paste it into chats, logs, screenshots, tutorials, or public repositories. Confirm the key is fully masked before taking any screenshot. If you notice unusual usage, revoke and recreate the key in the provider console immediately.
4. Enable Image Input
Make one check first: confirm in the provider's official model list that this model supports image input. Only after it is confirmed, enable Image as follows; if you cannot confirm it, treat the model as text-only and keep only Text enabled — do not guess from the model's family name.
When editing deepseek-v4-flash-vision-exp:
- Select Text and Image under input types;
- Select Text under output types;
- Click Save.
Without the Image option, an uploaded screenshot may not be passed to the model as visual input.
5. Enable and Test the Model
- Return to the DeepSeek provider page and confirm that it shows Enabled;
- Create a new task;
- Select
DeepSeek/deepseek-v4-flash-vision-expin the model menu; - Upload an ordinary test image and ask a question about it;
- Treat the setup as complete only when the image is described correctly.
If the test fails, or you see errors such as "model not found" or "provider rejected the request", troubleshoot in this order:
- Go back to the provider console and copy the real model ID again; the case and symbols must match exactly;
- Check that the API format and Base URL belong to the same protocol (see the rules below);
- Verify that the API key is valid, along with your balance or quota, service region, and plan;
- Select the model directly in a normal conversation and send a text-only test message (no image), to tell a model configuration problem apart from an operating mistake in the task.
If all of the above look fine but the test still fails, pause: confirm which API format the model actually supports in the provider's documentation, then reselect the format and recheck the model ID accordingly. Do not keep blindly trying different combinations.
6. Add Another Provider
Other providers use the same information pattern:
Base URL
+ API format
+ API Key
+ model ID
+ input capabilities (text / image / video)
Do not copy DeepSeek's URL or model ID into another provider. Use the API details supplied by that provider.
Popular Chinese Provider Cheat Sheet
How to choose before reading the table:
- Whichever provider you already have an API key for, look only at that provider's row; do not configure several at once;
- If you are unsure which format to pick, prefer Chat Completions — it is the common format that every provider's official documentation clearly supports; if the provider's documentation explicitly requires another format for a model, follow the documentation;
Chat Completions,Responses, andAnthropic Messagesare three wire-level API formats, not model names. The same provider may use a different Base URL under a different format, so switching formats means switching to the matching URL.
The table below is a cheat sheet for the Chat Completions format (addresses verified reachable on September 8, 2026). Endpoints may change, so always follow the provider's current official documentation.
| Provider | Base URL | Best for |
|---|---|---|
| Zhipu GLM (BigModel) | https://open.bigmodel.cn/api/paas/v4/ | Chinese tasks, daily workhorse |
| DeepSeek | https://api.deepseek.com | Strong reasoning, batch text |
| Kimi (Moonshot AI) | https://api.moonshot.cn/v1 | Long documents, large context |
| Alibaba Cloud Bailian (Tongyi Qwen) | https://dashscope.aliyuncs.com/compatible-mode/v1 | Qwen models, multimodal vision |
| SiliconFlow | https://api.siliconflow.cn/v1 | One key for many Chinese models |
- The API format and Base URL must come from the same row. Do not enter an Anthropic URL under the Chat Completions format, and do not edit addresses on a hunch.
- Copy addresses exactly as written, including
/v1and trailing slashes. Do not add/v1to DeepSeek or remove it from Kimi — each provider designs its URLs differently, so they cannot be mixed. - Copy model IDs from the provider console. Case and symbols must match exactly; never type a model ID from memory if it is not in the console list.
Zhipu, DeepSeek, and Kimi also offer the Anthropic Messages format, at https://open.bigmodel.cn/api/anthropic, https://api.deepseek.com/anthropic, and https://api.moonshot.cn/anthropic respectively. Use them only when the official documentation clearly supports them and you truly need them; otherwise stay on Chat Completions.
Provider-Specific Notes
Zhipu GLM (BigModel) · The provider behind the built-in plan; the workhorse for Chinese tasks, coding, and everyday work.
- API key: open.bigmodel.cn/usercenter/apikeys
- Model list: docs.bigmodel.cn model overview
- Note: this table lists Chat Completions only; for other formats, follow Zhipu's current official documentation.
DeepSeek · Strong reasoning, batch text, and cost-sensitive tasks; this guide used it for the full walkthrough (the Responses format in the walkthrough above is also one of DeepSeek's supported formats).
- API key: platform.deepseek.com/api_keys
- Documentation: api-docs.deepseek.com
- Note: Responses is a newer format, so rely on the official documentation for which parameters it supports; for vision tasks, pick only models officially marked as image-input capable.
Kimi / Moonshot AI · Long documents, knowledge work, and large repository understanding.
- API key: platform.kimi.com/console/api-keys
- Model list: platform.kimi.com/docs/models
- Note: the same model may expose different image and tool-calling capabilities under different formats; check the documentation for the format you choose.
Alibaba Cloud Bailian (Tongyi Qwen) · When you need multimodal models such as Qwen-VL.
- API key: sign in to the Bailian console and create one on the API-KEY management page
- Note: Bailian's endpoint depends on the region your account activated, so follow what the console and official documentation currently show; the URL above is the general OpenAI-compatible entry.
SiliconFlow · When you want one key to compare many Chinese models.
- API key: cloud.siliconflow.cn/account/ak
- Model square: cloud.siliconflow.cn/models
- Note: models on the platform come and go, so verify vision and tool-calling support model by model.
Overseas models (such as Google Gemini) are outside the scope of this table; if you need one, follow the current official Google AI Studio documentation.
Completion Check
Built-in plan only
- A working BigModel plan is selected;
-
GLM-5.3-Flashis selected for a new task.
These two items are enough to begin real work.
A third-party model was also added
- The provider shows Enabled;
- The API format and Base URL come from the same cheat-sheet row (or the same place in the provider's documentation);
- The model ID was copied from the provider console;
- The model can be selected in a new task;
- The text test succeeds;
- Image input is enabled for a vision model and the image test succeeds (text-only models do not need Image enabled).
Next Step
After configuring your models, continue to Step 4 · Install the ZCode AI Agent Team so different models can collaborate through specialist roles.