feat: Added support for modifying the reasoning effort of reasoning models

src/client/bedrock.rs

@@ -325,6 +325,7 @@ fn build_chat_completions_body(data: ChatCompletionsData, model: &Model) -> Resu
         mut messages,
         temperature,
         top_p,
+        reasoning_effort,
         functions,
         stream: _,
     } = data;
@@ -457,6 +458,9 @@ fn build_chat_completions_body(data: ChatCompletionsData, model: &Model) -> Resu
     if let Some(v) = top_p {
         body["inferenceConfig"]["topP"] = v.into();
     }
+    if let Some(v) = reasoning_effort {
+        body["additionalModelRequestFields"] = json!({ "output_config": { "effort": v } });
+    }
     if let Some(functions) = functions {
         let tools: Vec<_> = functions
             .iter()

src/client/claude.rs

@@ -251,6 +251,7 @@ pub fn claude_build_chat_completions_body(
         mut messages,
         temperature,
         top_p,
+        reasoning_effort,
         functions,
         stream,
     } = data;
@@ -369,6 +370,9 @@ pub fn claude_build_chat_completions_body(
     if let Some(v) = top_p {
         body["top_p"] = v.into();
     }
+    if let Some(v) = reasoning_effort {
+        body["output_config"] = json!({ "effort": v });
+    }
     if stream {
         body["stream"] = true.into();
     }

src/client/common.rs

@@ -286,6 +286,7 @@ pub struct ChatCompletionsData {
     pub messages: Vec<Message>,
     pub temperature: Option<f64>,
     pub top_p: Option<f64>,
+    pub reasoning_effort: Option<String>,
     pub functions: Option<Vec<FunctionDeclaration>>,
     pub stream: bool,
 }

src/client/model.rs

@@ -289,6 +289,14 @@ impl Model {
         }
         Ok(())
     }
+
+    pub fn reasoning_levels(&self) -> &[String] {
+        &self.data.reasoning_levels
+    }
+
+    pub fn default_reasoning_effort(&self) -> Option<&str> {
+        self.data.default_reasoning_effort.as_deref()
+    }
 }
 
 #[derive(Debug, Clone, Default, Serialize, Deserialize)]
@@ -316,6 +324,10 @@ pub struct ModelData {
     pub supports_vision: bool,
     #[serde(default, skip_serializing_if = "std::ops::Not::not")]
     pub supports_function_calling: bool,
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub reasoning_levels: Vec<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub default_reasoning_effort: Option<String>,
     #[serde(default, skip_serializing_if = "std::ops::Not::not")]
     no_stream: bool,
     #[serde(default, skip_serializing_if = "std::ops::Not::not")]

src/client/openai.rs

@@ -356,6 +356,7 @@ pub fn openai_build_chat_completions_body(data: ChatCompletionsData, model: &Mod
         messages,
         temperature,
         top_p,
+        reasoning_effort,
         functions,
         stream,
     } = data;
@@ -454,6 +455,9 @@ pub fn openai_build_chat_completions_body(data: ChatCompletionsData, model: &Mod
     if let Some(v) = top_p {
         body["top_p"] = v.into();
     }
+    if let Some(v) = reasoning_effort {
+        body["reasoning_effort"] = v.into();
+    }
     if stream {
         body["stream"] = true.into();
     }
@@ -534,6 +538,7 @@ pub fn openai_build_responses_body(data: ChatCompletionsData, model: &Model) ->
         messages,
         temperature,
         top_p,
+        reasoning_effort,
         functions,
         stream,
     } = data;
@@ -590,6 +595,9 @@ pub fn openai_build_responses_body(data: ChatCompletionsData, model: &Model) ->
     if let Some(v) = top_p {
         body["top_p"] = v.into();
     }
+    if let Some(v) = reasoning_effort {
+        body["reasoning"] = json!({ "effort": v });
+    }
     if stream {
         body["stream"] = true.into();
     }

src/client/vertexai.rs

@@ -334,6 +334,7 @@ pub fn gemini_build_chat_completions_body(
         mut messages,
         temperature,
         top_p,
+        reasoning_effort,
         functions,
         stream: _,
     } = data;
@@ -426,6 +427,9 @@ pub fn gemini_build_chat_completions_body(
     if let Some(v) = top_p {
         body["generationConfig"]["topP"] = v.into();
     }
+    if let Some(v) = reasoning_effort {
+        body["generation_config"]["thinking_level"] = v.into();
+    }
 
     if let Some(functions) = functions {
         // Gemini doesn't support functions with parameters that have empty properties, so we need to patch it.