How to Select a Model for the Agent
You can configure the AI model your agent will use and its response generation behavior from the Model Settings section.
This section includes the Model, Temperature, and Maximum Tokens settings.
You can change these settings manually from the agent creation screen or update them through SkyBuilder.
Structure of This Section
| Subsection | Content |
|---|---|
| 3.1 Model Selection | How to select a model and switch models during a chat |
| 3.2 Temperature | Balance between response consistency and variation |
| 3.3 Maximum Tokens | Response length limit |
| 3.4 Changing Model Settings with SkyBuilder | Updating settings using natural language |
| 3.5 Which Model Should I Choose? | Reference to the LLM Selection Guide |
3.1 Model Selection
Use the Model field to select the AI model your agent will use. From the dropdown menu, you can choose the model that best matches the intended use of your agent from the models available in SkyStudio.
When selecting a model, it is recommended to consider the tasks the agent will perform. Different models may be more suitable for different use cases such as data analysis, content generation, technical tasks, or general question answering.
Switching Models During a Chat
When starting a new conversation, you can select the model you want to use from the model menu. You can also switch to another model after the conversation has started; the existing context, including chat history, documents, text, and websites, is transferred to the newly selected model without losing the current conversation context.
| Topic | Behavior |
|---|---|
| Switching during a chat | Supported; the existing context is transferred to the newly selected model |
| Data loss | No conversation context is lost when switching |
| Data retention | User data is not stored by the models; each request receives the required context at the time of execution |
| Default model | The preferred default model can be configured from account settings |
Note: If you are unsure which model to choose, see the dedicated LLM Selection Guide.
3.2 Temperature
Temperature determines how consistent or varied the agent's responses will be. The value can be configured between 0 and 1.
| Value Range | Result | Suitable For |
|---|---|---|
| Low values | More consistent and predictable responses | Use cases where consistency is important |
| Medium values | Balanced responses | General use |
| High values | More creative and varied responses | Use cases where variation is desired |
A label such as Precise may be displayed next to the slider to indicate the response style associated with the selected temperature level.
3.3 Maximum Tokens
Maximum Tokens defines the maximum amount of content the agent can generate in a single response.
A higher value can be preferred for agents expected to produce long and comprehensive responses, while a lower value may be more suitable for agents designed to provide short and direct answers.
The maximum token value available may vary depending on the selected model.
3.4 Changing Model Settings with SkyBuilder
The Model, Temperature, and Maximum Tokens settings can also be changed through SkyBuilder. You can describe the changes you want using natural language, and SkyBuilder can update the relevant settings accordingly.
For example, you can ask SkyBuilder to select a different model, reduce the temperature for more consistent responses, or increase the maximum token value for longer outputs.
You can review the changes made by SkyBuilder in the Model Settings section and manually adjust them when necessary.
3.5 Which Model Should I Choose?
A separate reference document is available to help you identify the most suitable model for your use case.
The guide includes:
| Section | Content |
|---|---|
| Quick Selection Table | Model recommendations based on priorities such as accuracy, speed, cost, long context, and multimodality |
| Skymod — GOAT Thinking | Model designed for scenarios where data remains within organizational boundaries and KVKK/GDPR compliance is required |
| OpenAI Models | GPT-5.6 family, GPT-5 family, GPT-4.1, and o-series models |
| Anthropic Models | Claude 5 Sonnet, Opus, and Sonnet variants |
| Gemini Models | Gemini 3.1 Pro, 3 Flash, 2.5 Pro, and 2.5 Flash |
For each model, the guide provides separate sections for Use Cases, Task Suitability, and, where applicable, Our Recommendations.