diff --git a/config.template.toml b/config.template.toml
index e40bfc4d4d..1b6cd11c9a 100644
--- a/config.template.toml
+++ b/config.template.toml
@@ -200,6 +200,23 @@ model = "gpt-4o"
 # https://github.com/All-Hands-AI/OpenHands/pull/4711
 #native_tool_calling = None
 
+# Safety settings for models that support them (e.g., Mistral AI, Gemini)
+# Example for Mistral AI:
+# safety_settings = [
+#   { "category" = "hate", "threshold" = "low" },
+#   { "category" = "harassment", "threshold" = "low" },
+#   { "category" = "sexual", "threshold" = "low" },
+#   { "category" = "dangerous", "threshold" = "low" }
+# ]
+#
+# Example for Gemini:
+# safety_settings = [
+#   { "category" = "HARM_CATEGORY_HARASSMENT", "threshold" = "BLOCK_NONE" },
+#   { "category" = "HARM_CATEGORY_HATE_SPEECH", "threshold" = "BLOCK_NONE" },
+#   { "category" = "HARM_CATEGORY_SEXUALLY_EXPLICIT", "threshold" = "BLOCK_NONE" },
+#   { "category" = "HARM_CATEGORY_DANGEROUS_CONTENT", "threshold" = "BLOCK_NONE" }
+# ]
+#safety_settings = []
 
 
 [llm.gpt4o-mini]
diff --git a/docs/usage/llms/llms.mdx b/docs/usage/llms/llms.mdx
index 8b7d6b1859..8dea4beffb 100644
--- a/docs/usage/llms/llms.mdx
+++ b/docs/usage/llms/llms.mdx
@@ -73,6 +73,15 @@ We have a few guides for running OpenHands with specific model providers:
 - [OpenAI](/usage/llms/openai-llms)
 - [OpenRouter](/usage/llms/openrouter)
 
+## Model Customization
+
+LLM providers have specific settings that can be customized to optimize their performance with OpenHands, such as:
+
+- **Custom Tokenizers**: For specialized models, you can add a suitable tokenizer
+- **Native Tool Calling**: Toggle native function/tool calling capabilities
+
+For detailed information about model customization, see [LLM Configuration Options](configuration-options#llm-customization).
+
 ### API retries and rate limits
 
 LLM providers typically have rate limits, sometimes very low, and may require retries. OpenHands will automatically
diff --git a/openhands/core/config/llm_config.py b/openhands/core/config/llm_config.py
index 6a3fe10190..c39986746a 100644
--- a/openhands/core/config/llm_config.py
+++ b/openhands/core/config/llm_config.py
@@ -45,6 +45,7 @@ class LLMConfig(BaseModel):
         native_tool_calling: Whether to use native tool calling if supported by the model. Can be True, False, or not set.
         reasoning_effort: The effort to put into reasoning. This is a string that can be one of 'low', 'medium', 'high', or 'none'. Exclusive for o1 models.
         seed: The seed to use for the LLM.
+        safety_settings: Safety settings for models that support them (like Mistral AI and Gemini).
     """
 
     model: str = Field(default='claude-sonnet-4-20250514')
@@ -86,6 +87,10 @@ class LLMConfig(BaseModel):
     native_tool_calling: bool | None = Field(default=None)
     reasoning_effort: str | None = Field(default='high')
     seed: int | None = Field(default=None)
+    safety_settings: list[dict[str, str]] | None = Field(
+        default=None,
+        description='Safety settings for models that support them (like Mistral AI and Gemini)',
+    )
 
     model_config = {'extra': 'forbid'}
 
diff --git a/openhands/llm/llm.py b/openhands/llm/llm.py
index 91e0c79f15..1fddb51218 100644
--- a/openhands/llm/llm.py
+++ b/openhands/llm/llm.py
@@ -182,6 +182,12 @@ class LLM(RetryMixin, DebugMixin):
             kwargs['max_tokens'] = self.config.max_output_tokens
             kwargs.pop('max_completion_tokens')
 
+        # Add safety settings for models that support them
+        if 'mistral' in self.config.model.lower() and self.config.safety_settings:
+            kwargs['safety_settings'] = self.config.safety_settings
+        elif 'gemini' in self.config.model.lower() and self.config.safety_settings:
+            kwargs['safety_settings'] = self.config.safety_settings
+
         self._completion = partial(
             litellm_completion,
             model=self.config.model,