Skip to main content
Updated in 2026.19 The table provides an overview of the LLM providers supported by Cognigy, detailing the standard models and their compatibility with Cognigy features. You can also add a custom model that isn’t listed in the table. To do so, select the Custom Model parameter when creating a model and specify both the model type and name. For more information about adding a custom or standard model for a selected LLM provider, refer to the All LLM Providers page.

​
More Information

1 Design-time features include Intent Sentence Generation, Flow Generation, Adaptive Card Generation, Lexicon Generation.
2 Reasoning models consume more tokens and may incur higher costs. The models are optimized for tasks that require complex problem-solving and logical reasoning. Before using these models in production, test token consumption in debug mode and use them with caution. To reduce costs, consider using a non-reasoning model. For more information about reasoning models, refer to the Microsoft Azure OpenAI, OpenAI, and Google documentation.
Anthropic’s claude-sonnet-4-6 model doesn’t support assistant message prefilling. To keep your Flows compatible, ensure that the last transcript step before the LLM call is a user message. For example, add an Add Transcript Step Node before the LLM call and set the role to user.
3 For Knowledge AI, we recommend using text-embedding-ada-002. However, if you want to use text-embedding-3-small and text-embedding-3-large, make sure that you familiarize yourself with the restrictions of these models in Which Model to Choose?.
4 The *-latest suffix indicates that the model you select in Cognigy points to the latest version of the model. For more information, read Anthropic’s or Mistral AI’s models documentation.
5 For Cognigy 2025.10 and earlier versions, the option to select this model is hidden behind the FEATURE_ENABLE_AWS_BEDROCK_EMBEDDING_LLM_WHITELIST feature flag.
6 Note that some models from the Converse API might not support the AI Agent Node feature.
7 This model supports only the Responses API. When you select this model, the Responses API is selected by default, and the API Type field is locked.
8 The Conversation Analyzer uses the selected model to analyze session transcripts in Cognigy Insights. If no model is explicitly selected in Manage > Settings > Generative AI Settings, the Project’s default LLM is used.
9 These models don’t support token-sampling parameters such as temperature, top_p, and top_k. If you configure these parameters at the Node level or in Custom Model Options, they are automatically removed before the request is sent. To control model behavior, use the available thinking and budget settings instead.
10 The gemini-2.5-flash-lite and gemini-3.5-flash-lite models can return unstable results in the LLM Prompt Node. Test your Flows with these models before you use them in production.
11 This model defaults to reasoning.effort: "low" on the Responses API. The reasoning.effort parameter controls how much effort the model uses to think through a request. Higher effort can help the model handle more complex requests, while lower effort can yield faster responses. Supported values, such as "none", "low", "medium", and "high", and their behavior may vary by model and API. Check your LLM provider’s documentation for supported values and behavior. To override this default, set reasoning.effort in Custom Model Options. The default temperature is 1.0. To override this value, you must set reasoning.effort: "none" and adjust the Temperature parameter or set temperature in Custom Model Options.
Last modified on September 15, 2026