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

models.yaml

@@ -10,6 +10,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
     - name: gpt-5.6-terra
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -17,6 +19,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
     - name: gpt-5.6-luna
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -24,6 +28,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
     - name: gpt-5.5
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -31,6 +37,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: gpt-5.5-pro
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -38,6 +46,8 @@
       output_price: 180
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: high
     - name: gpt-5.4
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -45,6 +55,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: gpt-5.4-pro
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -52,6 +64,8 @@
       output_price: 180
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: gpt-5.4-mini
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -59,6 +73,8 @@
       output_price: 4.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: gpt-5.4-nano
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -66,6 +82,8 @@
       output_price: 1.25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: gpt-5.3-codex
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -73,6 +91,8 @@
       output_price: 14
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: chat-latest
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -87,6 +107,17 @@
       output_price: 14
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
+    - name: gpt-5.2-pro
+      max_input_tokens: 400000
+      max_output_tokens: 128000
+      input_price: 21
+      output_price: 168
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: gpt-5.1
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -94,6 +125,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high]
+      default_reasoning_effort: none
     - name: gpt-5.1-chat-latest
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -101,6 +134,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high]
+      default_reasoning_effort: none
     - name: gpt-5
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -108,6 +143,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: gpt-5-chat-latest
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -115,6 +152,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: gpt-5-mini
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -172,6 +211,8 @@
       supports_vision: true
       supports_function_calling: true
       system_prompt_prefix: Formatting re-enabled
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: medium
       patch:
         body:
           max_tokens: null
@@ -285,18 +326,24 @@
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: gemini-3-flash-preview
       max_input_tokens: 1048576
       max_output_tokens: 65536
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-3.1-flash-lite
       max_input_tokens: 1048576
       max_output_tokens: 65536
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: minimal
     - name: gemini-3.1-pro-preview
       max_input_tokens: 1048576
       max_output_tokens: 65535
@@ -304,6 +351,8 @@
       output_price: 2.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-2.5-flash
       max_input_tokens: 1048576
       max_output_tokens: 65536
@@ -311,6 +360,8 @@
       output_price: 0
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: low
     - name: gemini-2.5-pro
       max_input_tokens: 1048576
       max_output_tokens: 65536
@@ -318,6 +369,8 @@
       output_price: 0
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-2.5-flash-lite
       max_input_tokens: 1000000
       max_output_tokens: 64000
@@ -329,10 +382,14 @@
       max_input_tokens: 1048576
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, high]
+      default_reasoning_level: high
     - name: gemini-3-flash-preview
       max_input_tokens: 1048576
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_level: high
     - name: gemma-3-27b-it
       max_input_tokens: 131072
       max_output_tokens: 8192
@@ -358,6 +415,8 @@
       output_price: 50
       supports_function_calling: true
       supports_vision: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-8
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -366,6 +425,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-7
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -374,6 +435,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -382,6 +445,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-6:thinking
       real_name: claude-opus-4-6
       max_input_tokens: 200000
@@ -406,6 +471,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-sonnet-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -414,6 +481,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: claude-sonnet-4-6:thinking
       real_name: claude-sonnet-4-6
       max_input_tokens: 200000
@@ -856,18 +925,24 @@
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: gemini-3-flash-preview
       max_input_tokens: 1048576
       max_output_tokens: 65536
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-3.1-flash-lite
       max_input_tokens: 1048576
       max_output_tokens: 65536
       input_price: 0.2
       output_price: 1.5
       supports_function_calling: true
+      reasoning_levels: [minimal, high]
+      default_reasoning_effort: minimal
     - name: gemini-3.1-pro-preview
       max_input_tokens: 1048576
       max_output_tokens: 65536
@@ -875,6 +950,8 @@
       output_price: 12
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-2.5-flash
       max_input_tokens: 1048576
       max_output_tokens: 65535
@@ -882,6 +959,8 @@
       output_price: 2.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: medium
     - name: gemini-2.5-pro
       max_input_tokens: 1048576
       max_output_tokens: 65536
@@ -889,6 +968,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: gemini-2.5-flash-lite
       max_input_tokens: 1048576
       max_output_tokens: 65536
@@ -900,10 +981,14 @@
       max_input_tokens: 1048576
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, high]
+      default_reasoning_effort: high
     - name: gemini-3-flash-preview
       max_input_tokens: 1048576
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: high
     - name: claude-fable-5
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -912,6 +997,8 @@
       output_price: 50
       supports_function_calling: true
       supports_vision: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-8
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -920,6 +1007,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-7
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -928,6 +1017,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-opus-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -959,6 +1050,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: claude-sonnet-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -967,6 +1060,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: claude-sonnet-4-6:thinking
       real_name: claude-sonnet-4-6
       max_input_tokens: 200000
@@ -1099,6 +1194,8 @@
       output_price: 50
       supports_function_calling: true
       supports_vision: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-opus-4-8
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -1107,6 +1204,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-opus-4-7
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -1115,6 +1214,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-opus-4-6-v1
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -1123,6 +1224,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-opus-4-6-v1:thinking
       real_name: us.anthropic.claude-opus-4-6-v1
       max_input_tokens: 200000
@@ -1148,6 +1251,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-sonnet-4-6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -1156,6 +1261,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: us.anthropic.claude-sonnet-4-6:thinking
       real_name: us.anthropic.claude-sonnet-4-6
       max_input_tokens: 200000
@@ -1672,6 +1779,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.6-terra
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1679,6 +1788,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.6-luna
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1686,6 +1797,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh, max]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.5
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1693,6 +1806,8 @@
       output_price: 30
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.5-pro
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1700,6 +1815,8 @@
       output_price: 180
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: high
     - name: openai/gpt-5.4
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1707,6 +1824,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: openai/gpt-5.4-pro
       max_input_tokens: 1050000
       max_output_tokens: 128000
@@ -1714,6 +1833,8 @@
       output_price: 180
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.4-mini
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1721,6 +1842,8 @@
       output_price: 4.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: openai/gpt-5.4-nano
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1728,6 +1851,8 @@
       output_price: 1.25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
     - name: openai/gpt-5.3-codex
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1735,6 +1860,8 @@
       output_price: 14
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: openai/gpt-5.2
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1742,6 +1869,17 @@
       output_price: 14
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [none, low, medium, high, xhigh]
+      default_reasoning_effort: none
+    - name: openai/gpt-5.2-pro
+      max_input_tokens: 400000
+      max_output_tokens: 128000
+      input_price: 21
+      output_price: 168
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [medium, high, xhigh]
+      default_reasoning_effort: medium
     - name: openai/gpt-5
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1749,6 +1887,8 @@
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
     - name: openai/gpt-5-mini
       max_input_tokens: 400000
       max_output_tokens: 128000
@@ -1786,18 +1926,67 @@
       input_price: 0.04
       output_price: 0.16
       supports_function_calling: true
+    - name: google/gemini-3.5-flash
+      max_input_tokens: 1048576
+      max_output_tokens: 65536
+      input_price: 0.2
+      output_price: 1.5
+      supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: medium
+    - name: google/gemini-3-flash-preview
+      max_input_tokens: 1048576
+      max_output_tokens: 65536
+      input_price: 0.2
+      output_price: 1.5
+      supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: high
+    - name: google/gemini-3.1-flash-lite
+      max_input_tokens: 1048576
+      max_output_tokens: 65536
+      input_price: 0.2
+      output_price: 1.5
+      supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_effort: minimal
+    - name: google/gemini-3.1-pro-preview
+      max_input_tokens: 1048576
+      max_output_tokens: 65535
+      input_price: 0.3
+      output_price: 2.5
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
+    - name: google/gemini-3-pro-preview
+      max_input_tokens: 1048576
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [low, high]
+      default_reasoning_level: high
+    - name: google/gemini-3-flash-preview
+      max_input_tokens: 1048576
+      supports_vision: true
+      supports_function_calling: true
+      reasoning_levels: [minimal, low, medium, high]
+      default_reasoning_level: high
     - name: google/gemini-2.5-flash
       max_input_tokens: 1048576
       input_price: 0.3
       output_price: 2.5
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: low
     - name: google/gemini-2.5-pro
       max_input_tokens: 1048576
       input_price: 1.25
       output_price: 10
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high]
+      default_reasoning_effort: high
     - name: google/gemini-2.5-flash-lite
       max_input_tokens: 1048576
       input_price: 0.3
@@ -1827,6 +2016,8 @@
       output_price: 50
       supports_function_calling: true
       supports_vision: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-opus-4-8
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -1835,6 +2026,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-opus-4-7
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -1843,6 +2036,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-opus-4.6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -1851,6 +2046,8 @@
       output_price: 25
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-sonnet-5
       max_input_tokens: 1000000
       max_output_tokens: 128000
@@ -1859,6 +2056,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, xhigh, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-sonnet-4.6
       max_input_tokens: 200000
       max_output_tokens: 8192
@@ -1867,6 +2066,8 @@
       output_price: 15
       supports_vision: true
       supports_function_calling: true
+      reasoning_levels: [low, medium, high, max]
+      default_reasoning_effort: high
     - name: anthropic/claude-opus-4.5
       max_input_tokens: 200000
       max_output_tokens: 8192

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.

src/config/agent.rs

@@ -575,6 +575,10 @@ impl RoleLike for Agent {
         self.config.top_p
     }
 
+    fn reasoning_effort(&self) -> Option<String> {
+        self.config.reasoning_effort.clone()
+    }
+
     fn enabled_tools(&self) -> Option<Vec<String>> {
         None
     }
@@ -596,6 +600,10 @@ impl RoleLike for Agent {
         self.config.top_p = value;
     }
 
+    fn set_reasoning_effort(&mut self, value: Option<String>) {
+        self.config.reasoning_effort = value;
+    }
+
     fn set_enabled_tools(&mut self, value: Option<Vec<String>>) {
         match value {
             Some(tools) => {
@@ -637,6 +645,8 @@ pub struct AgentConfig {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub top_p: Option<f64>,
     #[serde(skip_serializing_if = "Option::is_none")]
+    pub reasoning_effort: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
     pub agent_session: Option<String>,
     #[serde(default)]
     pub auto_continue: bool,
@@ -732,6 +742,7 @@ impl AgentConfig {
             model_id: graph.model.clone(),
             temperature: graph.temperature,
             top_p: graph.top_p,
+            reasoning_effort: graph.reasoning_effort.clone(),
             description: graph.description.clone(),
             global_tools: graph.global_tools.clone(),
             mcp_servers: graph.mcp_servers.clone(),
@@ -766,6 +777,9 @@ impl AgentConfig {
         if let Some(v) = read_env_value::<f64>(&with_prefix("top_p")) {
             self.top_p = v;
         }
+        if let Some(v) = read_env_value::<String>(&with_prefix("reasoning_effort")) {
+            self.reasoning_effort = v;
+        }
         if let Ok(v) = env::var(with_prefix("global_tools"))
             && let Ok(v) = serde_json::from_str(&v)
         {

src/config/app_config.rs

@@ -21,6 +21,7 @@ pub struct AppConfig {
     pub model_id: String,
     pub temperature: Option<f64>,
     pub top_p: Option<f64>,
+    pub reasoning_effort: Option<String>,
 
     pub dry_run: bool,
     pub stream: bool,
@@ -100,6 +101,7 @@ impl Default for AppConfig {
             model_id: Default::default(),
             temperature: None,
             top_p: None,
+            reasoning_effort: None,
 
             dry_run: false,
             stream: true,
@@ -177,6 +179,7 @@ impl AppConfig {
             model_id: config.model_id,
             temperature: config.temperature,
             top_p: config.top_p,
+            reasoning_effort: None,
 
             dry_run: config.dry_run,
             stream: config.stream,
@@ -426,6 +429,9 @@ impl AppConfig {
         if let Some(v) = super::read_env_value::<f64>(&get_env_name("top_p")) {
             self.top_p = v;
         }
+        if let Some(v) = super::read_env_value::<String>(&get_env_name("reasoning_effort")) {
+            self.reasoning_effort = v;
+        }
 
         if let Some(Some(v)) = super::read_env_bool(&get_env_name("dry_run")) {
             self.dry_run = v;

src/config/input.rs

@@ -253,6 +253,10 @@ impl Input {
         patch_messages(&mut messages, model);
         model.guard_max_input_tokens(&messages)?;
         let (temperature, top_p) = (self.role().temperature(), self.role().top_p());
+        let reasoning_effort = self
+            .role()
+            .reasoning_effort()
+            .or_else(|| model.default_reasoning_effort().map(|s| s.to_string()));
         let functions = if model.supports_function_calling() {
             let fns = self.functions.clone();
             if let Some(vec) = &fns {
@@ -268,6 +272,7 @@ impl Input {
             messages,
             temperature,
             top_p,
+            reasoning_effort,
             functions,
             stream,
         })

src/config/request_context.rs

@@ -969,6 +969,16 @@ impl RequestContext {
         }
     }
 
+    pub fn set_reasoning_effort_on_role_like(&mut self, value: Option<String>) -> bool {
+        match self.role_like_mut() {
+            Some(role_like) => {
+                role_like.set_reasoning_effort(value);
+                true
+            }
+            None => false,
+        }
+    }
+
     pub fn set_enabled_tools_on_role_like(&mut self, value: Option<Vec<String>>) -> bool {
         match self.role_like_mut() {
             Some(role_like) => {
@@ -1121,6 +1131,10 @@ impl RequestContext {
                 super::format_option_value(&role.temperature()),
             ),
             ("top_p", super::format_option_value(&role.top_p())),
+            (
+                "reasoning_effort",
+                super::format_option_value(&role.reasoning_effort()),
+            ),
             (
                 "enabled_tools",
                 super::format_option_value(&role.enabled_tools().map(|v| v.join(","))),
@@ -2009,6 +2023,12 @@ impl RequestContext {
                     self.update_app_config(|app| app.top_p = value);
                 }
             }
+            "reasoning_effort" => {
+                let value: Option<String> = super::parse_value(value)?;
+                if !self.set_reasoning_effort_on_role_like(value.clone()) {
+                    self.update_app_config(|app| app.reasoning_effort = value);
+                }
+            }
             "enabled_tools" => {
                 let raw: Option<String> = super::parse_value(value)?;
                 let parsed: Option<Vec<String>> = raw.map(|s| super::csv_to_vec(&s));
@@ -2303,6 +2323,10 @@ impl RequestContext {
                     super::map_completion_values(values)
                 }
                 ".macro" => super::map_completion_values(paths::list_macros()),
+                ".reasoning" => {
+                    let levels = self.current_model().reasoning_levels();
+                    levels.iter().map(|v| (v.clone(), None)).collect()
+                }
                 ".starter" => match &self.agent {
                     Some(agent) => agent
                         .conversation_starters()
@@ -2318,6 +2342,7 @@ impl RequestContext {
                         "continuation_prompt",
                         "temperature",
                         "top_p",
+                        "reasoning_effort",
                         "enabled_tools",
                         "enabled_mcp_servers",
                         "inject_todo_instructions",
@@ -2507,6 +2532,10 @@ impl RequestContext {
                 }
                 "skill_instructions" => vec!["null".to_string()],
                 "memory" => super::complete_bool(self.should_inject_memory()),
+                "reasoning_effort" => {
+                    let levels = self.current_model().reasoning_levels();
+                    levels.to_vec()
+                }
                 _ => vec![],
             };
             values = candidates.into_iter().map(|v| (v, None)).collect();

src/config/role.rs

@@ -32,7 +32,9 @@ pub trait RoleLike {
     fn enabled_mcp_servers(&self) -> Option<Vec<String>>;
     fn set_model(&mut self, model: Model);
     fn set_temperature(&mut self, value: Option<f64>);
+    fn reasoning_effort(&self) -> Option<String>;
     fn set_top_p(&mut self, value: Option<f64>);
+    fn set_reasoning_effort(&mut self, value: Option<String>);
     fn set_enabled_tools(&mut self, value: Option<Vec<String>>);
     fn set_enabled_mcp_servers(&mut self, value: Option<Vec<String>>);
 }
@@ -51,6 +53,8 @@ pub struct Role {
     temperature: Option<f64>,
     #[serde(skip_serializing_if = "Option::is_none")]
     top_p: Option<f64>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    reasoning_effort: Option<String>,
     #[serde(
         default,
         skip_serializing_if = "Option::is_none",
@@ -116,6 +120,9 @@ impl Role {
                     "model" => role.model_id = value.as_str().map(|v| v.to_string()),
                     "temperature" => role.temperature = value.as_f64(),
                     "top_p" => role.top_p = value.as_f64(),
+                    "reasoning_effort" => {
+                        role.reasoning_effort = value.as_str().map(|v| v.to_string())
+                    }
                     "enabled_tools" => role.enabled_tools = parse_string_or_array(value),
                     "enabled_mcp_servers" => {
                         role.enabled_mcp_servers = parse_string_or_array(value)
@@ -170,6 +177,9 @@ impl Role {
         if let Some(top_p) = self.top_p() {
             metadata.push(format!("top_p: {top_p}"));
         }
+        if let Some(reasoning_effort) = self.reasoning_effort() {
+            metadata.push(format!("reasoning_effort: {reasoning_effort}"));
+        }
         if let Some(enabled_tools) = &self.enabled_tools {
             let inline = serde_json::to_string(enabled_tools).unwrap_or_else(|_| "[]".to_string());
             metadata.push(format!("enabled_tools: {inline}"));
@@ -256,6 +266,9 @@ impl Role {
             enabled_tools,
             enabled_mcp_servers,
         );
+        if let Some(v) = role_like.reasoning_effort() {
+            self.set_reasoning_effort(Some(v));
+        }
     }
 
     pub fn batch_set(
@@ -410,6 +423,10 @@ impl RoleLike for Role {
         self.top_p
     }
 
+    fn reasoning_effort(&self) -> Option<String> {
+        self.reasoning_effort.clone()
+    }
+
     fn enabled_tools(&self) -> Option<Vec<String>> {
         self.enabled_tools.clone()
     }
@@ -433,6 +450,10 @@ impl RoleLike for Role {
         self.top_p = value;
     }
 
+    fn set_reasoning_effort(&mut self, value: Option<String>) {
+        self.reasoning_effort = value;
+    }
+
     fn set_enabled_tools(&mut self, value: Option<Vec<String>>) {
         self.enabled_tools = value;
     }

src/config/session.rs

@@ -24,6 +24,8 @@ pub struct Session {
     temperature: Option<f64>,
     #[serde(skip_serializing_if = "Option::is_none")]
     top_p: Option<f64>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    reasoning_effort: Option<String>,
     #[serde(
         default,
         skip_serializing_if = "Option::is_none",
@@ -401,6 +403,7 @@ impl Session {
         self.model_id = role.model().id();
         self.temperature = role.temperature();
         self.top_p = role.top_p();
+        self.reasoning_effort = role.reasoning_effort();
         self.enabled_tools = role.enabled_tools();
         self.enabled_mcp_servers = role.enabled_mcp_servers();
         self.model = role.model().clone();
@@ -792,6 +795,10 @@ impl RoleLike for Session {
         self.top_p
     }
 
+    fn reasoning_effort(&self) -> Option<String> {
+        self.reasoning_effort.clone()
+    }
+
     fn enabled_tools(&self) -> Option<Vec<String>> {
         self.enabled_tools.clone()
     }
@@ -823,6 +830,13 @@ impl RoleLike for Session {
         }
     }
 
+    fn set_reasoning_effort(&mut self, value: Option<String>) {
+        if self.reasoning_effort != value {
+            self.reasoning_effort = value;
+            self.dirty = true;
+        }
+    }
+
     fn set_enabled_tools(&mut self, value: Option<Vec<String>>) {
         if self.enabled_tools != value {
             self.enabled_tools = value;

src/graph/llm.rs

@@ -329,6 +329,9 @@ fn build_inline_role(
     if let Some(p) = node.top_p {
         role.set_top_p(Some(p));
     }
+    if let Some(v) = &node.reasoning_effort {
+        role.set_reasoning_effort(Some(v.clone()));
+    }
 
     if node.tools.as_deref().unwrap_or_default().is_empty() {
         role.set_enabled_tools(Some(Vec::new()));
@@ -499,6 +502,7 @@ mod tests {
             model: None,
             temperature: None,
             top_p: None,
+            reasoning_effort: None,
             fallback: None,
             max_attempts: 1,
             max_iterations: 10,

src/graph/types.rs

@@ -25,6 +25,9 @@ pub struct Graph {
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub top_p: Option<f64>,
 
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub reasoning_effort: Option<String>,
+
     #[serde(default)]
     pub global_tools: Vec<String>,
 
@@ -288,6 +291,9 @@ pub struct LlmNode {
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub top_p: Option<f64>,
 
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub reasoning_effort: Option<String>,
+
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub fallback: Option<String>,
 

src/graph/validator.rs

@@ -946,6 +946,7 @@ mod tests {
             model: None,
             temperature: None,
             top_p: None,
+            reasoning_effort: None,
             global_tools: Vec::new(),
             mcp_servers: Vec::new(),
             skills_enabled: None,
@@ -1048,6 +1049,7 @@ mod tests {
                 model: None,
                 temperature: None,
                 top_p: None,
+                reasoning_effort: None,
                 fallback: fallback.map(String::from),
                 max_attempts: 1,
                 max_iterations: 10,

src/rag/graph.rs

@@ -362,6 +362,7 @@ pub async fn extract_entities(
         messages,
         temperature: Some(0.0),
         top_p: None,
+        reasoning_effort: None,
         functions: None,
         stream: false,
     };

src/repl/mod.rs

@@ -52,7 +52,7 @@ pub const DEFAULT_CONTINUATION_PROMPT: &str = indoc! {"
     4. Continue with the next pending item now. Call tools immediately."
 };
 
-static REPL_COMMANDS: LazyLock<[ReplCommand; 51]> = LazyLock::new(|| {
+static REPL_COMMANDS: LazyLock<[ReplCommand; 52]> = LazyLock::new(|| {
     [
         ReplCommand::new(".help", "Show this help guide", AssertState::pass()),
         ReplCommand::new(".info", "Show system info", AssertState::pass()),
@@ -259,6 +259,11 @@ static REPL_COMMANDS: LazyLock<[ReplCommand; 51]> = LazyLock::new(|| {
         ),
         ReplCommand::new(".copy", "Copy last response", AssertState::pass()),
         ReplCommand::new(".set", "Modify runtime settings", AssertState::pass()),
+        ReplCommand::new(
+            ".reasoning",
+            "Set the reasoning effort level for the current model",
+            AssertState::pass(),
+        ),
         ReplCommand::new(
             ".delete",
             "Delete roles, sessions, RAGs, or agents",
@@ -1070,6 +1075,15 @@ pub async fn run_repl_command(
                     println!("Usage: .set <key> <value>...")
                 }
             },
+            ".reasoning" => match args {
+                Some(level) => {
+                    let set_args = format!("reasoning_effort {level}");
+                    ctx.update(&set_args, abort_signal).await?;
+                }
+                None => {
+                    println!("Usage: .reasoning <level>")
+                }
+            },
             ".delete" => match args {
                 Some(args) => {
                     ctx.delete(args)?;
@@ -1600,8 +1614,8 @@ mod tests {
     }
 
     #[test]
-    fn repl_commands_has_50_entries() {
+    fn repl_commands_has_52_entries() {
-        assert_eq!(REPL_COMMANDS.len(), 50);
+        assert_eq!(REPL_COMMANDS.len(), 52);
     }
 
     #[test]