Multi-model collaboration is an advanced capability available on the chat page. When you send a message, the system calls multiple AI models you selected in parallel, lets each model answer independently, and then uses a summary model to compare and evaluate the responses from all models before producing a comprehensive conclusion. This is suitable for complex question verification, multi-perspective comparison, reducing the risk of errors from a single model, and similar scenarios.
1. Feature Entry
- Go to the Chat page by clicking "Chat" in the left navigation bar.
- In the toolbar above the input box at the bottom, find the "Multi-model Collaboration" button. By default, it is shown in a light, disabled state.
- Click the button to open the Multi-model Collaboration configuration panel as a side panel.
Tip: When disabled, the button displays "Multi-model Collaboration". After it is enabled, it changes to "Multi-model Collaboration ยท Enabled" and shows a green label next to it indicating the number of participating models and the summary model.
2. Configure Collaboration Parameters
In the side panel that appears, configure the following settings in order:
1. Select Participating Models
- In the "Participating Models" list, select the models you want to participate in answering.
- You must select at least 2 models, and up to 5 models are supported.
- The bottom of the panel displays "Selected N / Maximum 5" in real time.
Warning: The more participating models you select, the higher the quota consumption for a single conversation, because each model is billed independently.
2. Select the Summary Model
- In the "Summary Model" dropdown, select one model to summarize the answers from all participating models.
- The summary model reads all participating model responses, compares and analyzes them, and then outputs a comprehensive conclusion.
- It is recommended to select a model with strong reasoning capability as the summary model to obtain a higher-quality summary.
3. Set the Timeout
- In the "Timeout Settings" dropdown, select the maximum waiting time for a single collaboration: 30 seconds / 60 seconds / 120 seconds / 180 seconds.
- Participating models that have not completed within this time will be interrupted, and the summary model will summarize based on the responses that were returned successfully.
4. Confirm and Enable
- After completing the settings above, click the "Confirm and Enable" button in the lower-right corner of the panel.
- The panel closes, and the toolbar button changes to "Multi-model Collaboration ยท Enabled", indicating that collaboration mode has been activated.
3. Send a Collaboration Message
After multi-model collaboration is enabled, enter your question in the input box and send it as usual. The system will automatically run the collaboration workflow:
- Parallel answering: All participating models you selected start answering at the same time. Each model's response is displayed in real time in the conversation area as a separate streaming message card. The bottom of each card is labeled "Participating model: XXX".
- Real-time viewing: While the participating models are answering, you can view the output progress of each model in real time.
- Summary aggregation: After all participating models finish answering, or when the timeout is reached, the summary model starts automatically. It generates a structured comprehensive conclusion based on each model's response and displays it as a summary message card labeled "Collaboration summary: XXX".
Tip: By default, the summary includes modules such as question understanding, comparison of key points from each model, consensus and differences, comprehensive conclusion, and recommended actions, helping you quickly obtain the best answer.
4. Recommended Use Cases
| Scenario | Recommended | Description |
|---|---|---|
| Complex technical questions / professional domain verification | Strongly recommended | Cross-validate with multiple models to reduce the risk of hallucinations from a single model |
| Solution comparison / decision-making advice | Recommended | Obtain multi-perspective opinions and let the summary model weigh the pros and cons |
| Fact-checking / high data accuracy requirements | Recommended | Quickly identify reliable information through the consensus and differences module |
| Daily casual chat / single-turn Q&A | Not recommended | Parallel use of multiple models consumes more quota and is less cost-effective |
| Real-time scenarios sensitive to latency | Not recommended | Requires waiting for all participating models plus the summary model, resulting in longer response time |
Multi-model Collaboration Flowchart
Chat page toolbar
|
โโ Click "Multi-model Collaboration" โโโ Open configuration panel (popup)
|
โโ Select participating models (2-5 models)
โโ Select summary model (1 model)
โโ Set timeout
โโ Click "Confirm and Enable"
|
v
Toolbar displays "Enabled"
|
Enter a question in the input box and send
|
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
v v v
Participating model A (streaming) Participating model B (streaming) Participating model C (streaming)
Outputs answer in real time Outputs answer in real time Outputs answer in real time
| | |
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโผโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
v
All completed / timeout reached
|
v
Summary model reads each model's answer
|
v
Outputs comprehensive conclusion (comparison + consensus + suggestions)