> For the complete documentation index, see [llms.txt](https://docs-conversational-ai.syntphony.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs-conversational-ai.syntphony.com/language-models/advanced-model-settings.md). # Advanced Model Settings Fine-tune how language models reason and generate responses across Syntphony CAI. ## Overview **Advanced Model Settings** let you customize specific aspects of model behavior, such as **reasoning, response detail, randomness, vocabulary variation, repetition, and response limits**. The available settings depend on the **selected model and the Syntphony CAI feature being configured**. GPT-4 and GPT-5 family models expose different controls because they provide different ways to influence response generation. {% hint style="info" %} **Important:** Not every setting is available for every model or feature. Always configure the controls displayed for the selected model. {% endhint %} *** ## Settings by model family ### GPT-5 family GPT-5 family models introduce separate controls for **reasoning** and **response detail**: * **Reasoning Effort:** controls how much reasoning the model performs before generating a response. * **Verbosity:** controls the level of detail in the generated response. These settings work independently. For example, you can configure **higher reasoning with a concise response**, or **lower reasoning with a more detailed response**. {% hint style="info" %} **Note:** Available Reasoning Effort levels vary between GPT-5 family models. Only the values supported by the selected model are displayed. {% endhint %} #### Reasoning Effort **Reasoning Effort** controls how much reasoning the model performs before generating its response. Depending on the GPT-5 model, available levels may include **None, Minimal, Low, Medium, High, Xhigh, or Max**. Higher levels can be useful for tasks involving **multi-step reasoning, calculations, or ambiguity resolution**, but may also increase response time and token usage. Lower levels are more appropriate for simpler interactions where fast responses are preferred. {% hint style="info" %} **Best practice:** Higher is not always better. Select the level according to the complexity of the task and the expected balance between reasoning, latency, and consumption. {% endhint %} #### Verbosity **Verbosity** controls how much detail appears in the generated response: * **Low:** Concise responses with minimal detail. * **Medium:** Balanced responses with moderate detail. * **High:** More comprehensive and detailed responses. Low Verbosity can be useful for short conversational interactions, including chat and voice experiences. Higher levels can be appropriate when users need more detailed explanations. {% hint style="info" %} **Remember:** Reasoning Effort controls **how much the model reasons**; Verbosity controls **how much detail the user sees**. They are independent settings. {% endhint %} ### GPT-4 family GPT-4 family models provide controls focused on **generation variation and repetition**: * **Temperature:** controls randomness. Lower values produce more predictable responses, while higher values increase variation. * **Top P:** controls vocabulary variation by influencing the range of possible tokens. * **Frequency Penalty:** influences repetition according to how frequently tokens have already appeared. * **Presence Penalty:** influences whether previously used tokens are likely to appear again. {% hint style="info" %} **Tip:** Avoid treating Temperature and Top P as interchangeable controls. Adjust settings according to the specific response behavior you want to change. {% endhint %} #### Temperature **Temperature** controls the randomness of generated content. * **Lower values** produce more predictable and consistent responses. * **Higher values** increase variation and creativity. Use lower values when consistency is important and higher values when the interaction benefits from more varied responses. #### Top P **Top P** controls vocabulary variation by influencing the range of words the model considers when generating a response. * **Lower values** favor more common and predictable words or phrases. * **Higher values** allow a broader range of possible words. Use it to adjust how varied the model's vocabulary can be. #### Frequency Penalty **Frequency Penalty** controls repetition based on how frequently tokens have already appeared in the generated content. Higher values increase the influence of token frequency, helping reduce repeated words or expressions. #### Presence Penalty **Presence Penalty** considers whether a token has already appeared in the generated content, regardless of how frequently it was used. Higher values make the model more likely to introduce tokens that have not yet appeared, encouraging greater variation in the response. {% hint style="info" %} **Tip:** **Temperature** and **Top P** influence variation in different ways, while **Frequency Penalty** and **Presence Penalty** influence repetition. Adjust the setting that corresponds to the behavior you want to change rather than increasing multiple controls at once. {% endhint %} *** ## Understanding the settings ### Model-specific settings These settings control **how the selected language model generates responses**. Their availability and supported values depend on the model family and selected model.
SettingConfiguration goalAvailable values or control
Reasoning effortAdjust how much reasoning occurs before a response.Model-dependent levels
VerbosityAdjust the level of detail in a response.Low, Medium, High
TemperatureAdjust randomness in generated content.Numeric control or slider
Top PAdjust vocabulary variation.Numeric control or slider
Frequency PenaltyInfluence repetition according to token frequency.Numeric control or slider
Presence PenaltyInfluence whether previously used tokens appear again.Numeric control or slider
### General settings These settings control **how Syntphony CAI handles the model request or output** rather than the model's generation behavior itself.
SettingConfiguration goalControl
Response wait timeDefine how long Syntphony CAI waits for a model response.Slider, in seconds
Maximum TokensLimit the maximum amount of content generated in the output.Slider, in tokens
Restrict vocabularyPrevent specified words from appearing in an answer variation.Optional text field
Conversation historyControl how much previous conversation context is considered when generating an answer variation.Numeric control or slider
*** ## Settings by feature The controls available also depend on where the model is being used: | Feature | Family 4 | Family 5 | | ------------------------ | --------------------------------------------------------------------------- | ----------------------------------------------- | | AI Agents and Supervisor | Response wait time, Temperature, Top P, Presence penalty, Frequency penalty | Response wait time, Reasoning effort, Verbosity | | Prompt Cell | Maximum Tokens, Temperature, Top P, Presence penalty, Frequency penalty | Maximum Tokens, Reasoning effort, Verbosity | | Rephrase Answer | Temperature, Conversation history, Restrict vocabulary | Still not available on the platform | *** ## Configure advanced settings Configure the settings directly from the feature where the model is being used. For **AI Agents and Supervisor**, go to **Preferences** and expand **Advanced Settings**. The available controls automatically reflect the configured model.
AI Agent with model and general settings

AI Agent with model and general settings

For **Prompt Cells**, expand **Advanced Parameters** and configure the settings available for the selected model. Use **Generate Preview** to validate the combination of your Prompt and current parameters before saving.
Prompt Cell with model and general settings

Prompt Cell with model and general settings

For **Rephrase Answer**, enable the feature and configure the available generation control together with **Conversation History** and, when required, **Restrict Vocabulary**.
Rephrase Answer with model and general settings

Rephrase Answer with model and general settings

*** ## When changing models When switching between models or model families, review the available settings before publishing. Settings from one model family do not necessarily have direct equivalents in another. Instead of trying to reproduce the same parameter values, **test the resulting behavior and adjust the new model according to the expected outcome**. A simple workflow is: **Select model → Review available settings → Configure → Test → Adjust → Save or publish** {% hint style="info" %} **Best practice:** When changing model families, test representative conversations before publishing to validate response quality, reasoning, detail, and latency. {% endhint %} *** ## Troubleshooting
IssueCauseResolution
A Reasoning effort level is not availableSupported levels vary between GPT-5 models.Select one of the levels available for the selected model.
A previously available Reasoning effort level is unavailable after changing family 5 modelsSupported levels vary within family 5.Select a level available for the active model.
A setting is not displayedThe selected model or feature does not support it.Use the controls available for the active configuration.
--- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://docs-conversational-ai.syntphony.com/language-models/advanced-model-settings.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.