diff --git a/lib/req_llm/providers/azure.ex b/lib/req_llm/providers/azure.ex
index 719fc680..dc87c3f2 100644
--- a/lib/req_llm/providers/azure.ex
+++ b/lib/req_llm/providers/azure.ex
@@ -246,6 +246,11 @@ defmodule ReqLLM.Providers.Azure do
     max_completion_tokens: [
       type: :any,
       doc: "Maximum completion tokens (OpenAI reasoning models)"
+    ],
+    verbosity: [
+      type: {:or, [:atom, :string]},
+      doc:
+        "Constrains the verbosity of the model's response. Supported values: 'low', 'medium', 'high'. Defaults to 'medium'. (OpenAI models only)"
     ]
   ]
 
diff --git a/lib/req_llm/providers/azure/openai.ex b/lib/req_llm/providers/azure/openai.ex
index 8954d496..0bb54474 100644
--- a/lib/req_llm/providers/azure/openai.ex
+++ b/lib/req_llm/providers/azure/openai.ex
@@ -173,6 +173,7 @@ defmodule ReqLLM.Providers.Azure.OpenAI do
       |> maybe_put(:n, opts[:n])
       |> maybe_put(:reasoning_effort, provider_opts[:reasoning_effort])
       |> maybe_put(:service_tier, provider_opts[:service_tier])
+      |> add_verbosity(provider_opts)
       |> add_stream_options(opts)
       |> AdapterHelpers.add_parallel_tool_calls(opts, provider_opts)
       |> AdapterHelpers.translate_tool_choice_format()
@@ -205,6 +206,15 @@ defmodule ReqLLM.Providers.Azure.OpenAI do
     end
   end
 
+  defp add_verbosity(body, provider_opts) do
+    verbosity = provider_opts[:verbosity]
+    maybe_put(body, :verbosity, normalize_verbosity(verbosity))
+  end
+
+  defp normalize_verbosity(nil), do: nil
+  defp normalize_verbosity(v) when is_atom(v), do: Atom.to_string(v)
+  defp normalize_verbosity(v) when is_binary(v), do: v
+
   @doc """
   Formats an embedding request for Azure OpenAI.
 
diff --git a/lib/req_llm/providers/openai.ex b/lib/req_llm/providers/openai.ex
index 259a1eba..f0a75754 100644
--- a/lib/req_llm/providers/openai.ex
+++ b/lib/req_llm/providers/openai.ex
@@ -155,6 +155,11 @@ defmodule ReqLLM.Providers.OpenAI do
     service_tier: [
       type: {:or, [:atom, :string]},
       doc: "Service tier for request prioritization ('auto', 'default', 'flex' or 'priority')"
+    ],
+    verbosity: [
+      type: {:or, [:atom, :string]},
+      doc:
+        "Constrains the verbosity of the model's response. Supported values: 'low', 'medium', 'high'. Defaults to 'medium'."
     ]
   ]
 
diff --git a/lib/req_llm/providers/openai/chat_api.ex b/lib/req_llm/providers/openai/chat_api.ex
index 5917c5b5..cc29c591 100644
--- a/lib/req_llm/providers/openai/chat_api.ex
+++ b/lib/req_llm/providers/openai/chat_api.ex
@@ -117,6 +117,7 @@ defmodule ReqLLM.Providers.OpenAI.ChatAPI do
         |> add_stream_options(opts_map)
         |> add_reasoning_effort(opts_map)
         |> add_service_tier(opts_map)
+        |> add_verbosity(opts_map)
         |> add_response_format(opts_map)
         |> add_parallel_tool_calls(opts_map)
         |> translate_tool_choice_format()
@@ -207,6 +208,16 @@ defmodule ReqLLM.Providers.OpenAI.ChatAPI do
     maybe_put(body, :service_tier, service_tier)
   end
 
+  defp add_verbosity(body, request_options) do
+    provider_opts = request_options[:provider_options] || []
+    verbosity = provider_opts[:verbosity]
+    maybe_put(body, :verbosity, normalize_verbosity(verbosity))
+  end
+
+  defp normalize_verbosity(nil), do: nil
+  defp normalize_verbosity(v) when is_atom(v), do: Atom.to_string(v)
+  defp normalize_verbosity(v) when is_binary(v), do: v
+
   defp translate_tool_choice_format(body) do
     {tool_choice, body_key} =
       cond do
diff --git a/lib/req_llm/providers/openai/responses_api.ex b/lib/req_llm/providers/openai/responses_api.ex
index fb91a5c2..bac48f62 100644
--- a/lib/req_llm/providers/openai/responses_api.ex
+++ b/lib/req_llm/providers/openai/responses_api.ex
@@ -304,7 +304,7 @@ defmodule ReqLLM.Providers.OpenAI.ResponsesAPI do
     reasoning = encode_reasoning_effort(opts_map[:reasoning_effort])
     service_tier = opts_map[:service_tier] || provider_opts[:service_tier]
 
-    text_format = encode_text_format(provider_opts[:response_format])
+    text_format = encode_text_format(provider_opts[:response_format], provider_opts[:verbosity])
 
     final_input =
       if previous_response_id == nil and reasoning_items != [] do
@@ -768,34 +768,47 @@ defmodule ReqLLM.Providers.OpenAI.ResponsesAPI do
   defp encode_reasoning_effort(_), do: nil
 
   @doc false
-  def encode_text_format(nil), do: nil
+  def encode_text_format(response_format, verbosity \\ nil)
 
-  def encode_text_format(response_format) when is_map(response_format) do
-    # Extract type - could be atom or string key
+  def encode_text_format(nil, nil), do: nil
+
+  def encode_text_format(nil, verbosity) do
+    %{"verbosity" => normalize_verbosity(verbosity)}
+  end
+
+  def encode_text_format(response_format, verbosity) when is_map(response_format) do
     type = response_format[:type] || response_format["type"]
 
-    case type do
-      "json_schema" ->
-        json_schema = response_format[:json_schema] || response_format["json_schema"]
-        # Schema.to_json handles both keyword lists (converts) and maps (pass-through)
-        schema = ReqLLM.Schema.to_json(json_schema[:schema] || json_schema["schema"])
+    base =
+      case type do
+        "json_schema" ->
+          json_schema = response_format[:json_schema] || response_format["json_schema"]
+          schema = ReqLLM.Schema.to_json(json_schema[:schema] || json_schema["schema"])
 
-        # ResponsesAPI expects a flattened structure:
-        # text.format.{type, name, strict, schema} instead of text.format.json_schema.{name, strict, schema}
-        %{
-          "format" => %{
-            "type" => "json_schema",
-            "name" => json_schema[:name] || json_schema["name"],
-            "strict" => json_schema[:strict] || json_schema["strict"],
-            "schema" => schema
+          %{
+            "format" => %{
+              "type" => "json_schema",
+              "name" => json_schema[:name] || json_schema["name"],
+              "strict" => json_schema[:strict] || json_schema["strict"],
+              "schema" => schema
+            }
           }
-        }
 
-      _ ->
-        nil
+        _ ->
+          %{}
+      end
+
+    case {base, verbosity} do
+      {b, nil} when map_size(b) == 0 -> nil
+      {b, v} when map_size(b) == 0 -> %{"verbosity" => normalize_verbosity(v)}
+      {b, nil} -> b
+      {b, v} -> Map.put(b, "verbosity", normalize_verbosity(v))
     end
   end
 
+  defp normalize_verbosity(v) when is_atom(v), do: Atom.to_string(v)
+  defp normalize_verbosity(v) when is_binary(v), do: v
+
   defp decode_responses_success({req, resp}) do
     body = ReqLLM.Provider.Utils.ensure_parsed_body(resp.body)
 
diff --git a/test/provider/azure/azure_test.exs b/test/provider/azure/azure_test.exs
index f43f32ac..44dfa331 100644
--- a/test/provider/azure/azure_test.exs
+++ b/test/provider/azure/azure_test.exs
@@ -698,6 +698,45 @@ defmodule ReqLLM.Providers.AzureTest do
     end
   end
 
+  describe "verbosity option" do
+    test "OpenAI models include verbosity when provided as atom" do
+      context = ReqLLM.Context.new([ReqLLM.Context.user("Hello")])
+      opts = [stream: false, provider_options: [verbosity: :low]]
+
+      body = Azure.OpenAI.format_request("gpt-4o", context, opts)
+
+      assert body[:verbosity] == "low"
+    end
+
+    test "OpenAI models include verbosity when provided as string" do
+      context = ReqLLM.Context.new([ReqLLM.Context.user("Hello")])
+      opts = [stream: false, provider_options: [verbosity: "high"]]
+
+      body = Azure.OpenAI.format_request("gpt-4o", context, opts)
+
+      assert body[:verbosity] == "high"
+    end
+
+    test "OpenAI models omit verbosity when not provided" do
+      context = ReqLLM.Context.new([ReqLLM.Context.user("Hello")])
+      opts = [stream: false]
+
+      body = Azure.OpenAI.format_request("gpt-4o", context, opts)
+
+      refute Map.has_key?(body, :verbosity)
+    end
+
+    test "verbosity works with reasoning models" do
+      context = ReqLLM.Context.new([ReqLLM.Context.user("Hello")])
+      opts = [stream: false, provider_options: [verbosity: :medium, reasoning_effort: "high"]]
+
+      body = Azure.OpenAI.format_request("o3-mini", context, opts)
+
+      assert body[:verbosity] == "medium"
+      assert body[:reasoning_effort] == "high"
+    end
+  end
+
   describe "reasoning model features" do
     import ExUnit.CaptureLog
 
diff --git a/test/provider/openai/responses_api_unit_test.exs b/test/provider/openai/responses_api_unit_test.exs
index 63c85b27..f20261c0 100644
--- a/test/provider/openai/responses_api_unit_test.exs
+++ b/test/provider/openai/responses_api_unit_test.exs
@@ -308,6 +308,60 @@ defmodule Provider.OpenAI.ResponsesAPIUnitTest do
       assert body["text"]["format"]["strict"] == true
       assert body["text"]["format"]["schema"] == json_schema
     end
+
+    test "encodes verbosity when provided as atom" do
+      request = build_request(provider_options: [verbosity: :low])
+
+      encoded = ResponsesAPI.encode_body(request)
+      body = Jason.decode!(encoded.body)
+
+      assert body["text"]["verbosity"] == "low"
+    end
+
+    test "encodes verbosity when provided as string" do
+      request = build_request(provider_options: [verbosity: "high"])
+
+      encoded = ResponsesAPI.encode_body(request)
+      body = Jason.decode!(encoded.body)
+
+      assert body["text"]["verbosity"] == "high"
+    end
+
+    test "omits text field when no verbosity or response_format" do
+      request = build_request(provider_options: [])
+
+      encoded = ResponsesAPI.encode_body(request)
+      body = Jason.decode!(encoded.body)
+
+      refute Map.has_key?(body, "text")
+    end
+
+    test "encodes verbosity alongside response_format in text object" do
+      json_schema = %{
+        "type" => "object",
+        "properties" => %{"name" => %{"type" => "string"}},
+        "required" => ["name"]
+      }
+
+      response_format = %{
+        type: "json_schema",
+        json_schema: %{
+          name: "test_schema",
+          strict: true,
+          schema: json_schema
+        }
+      }
+
+      request =
+        build_request(provider_options: [response_format: response_format, verbosity: :medium])
+
+      encoded = ResponsesAPI.encode_body(request)
+      body = Jason.decode!(encoded.body)
+
+      assert body["text"]["format"]["type"] == "json_schema"
+      assert body["text"]["format"]["name"] == "test_schema"
+      assert body["text"]["verbosity"] == "medium"
+    end
   end
 
   describe "decode_response/1" do
diff --git a/test/providers/openai_test.exs b/test/providers/openai_test.exs
index 995acc85..187782cb 100644
--- a/test/providers/openai_test.exs
+++ b/test/providers/openai_test.exs
@@ -500,6 +500,59 @@ defmodule ReqLLM.Providers.OpenAITest do
 
       assert decoded["service_tier"] == "flex"
     end
+
+    test "encode_body includes verbosity when provided as atom" do
+      {:ok, model} = ReqLLM.model("openai:gpt-4o")
+      context = context_fixture()
+
+      mock_request = %Req.Request{
+        options: [
+          context: context,
+          model: model.model,
+          provider_options: [verbosity: :low]
+        ]
+      }
+
+      updated_request = OpenAI.encode_body(mock_request)
+      decoded = Jason.decode!(updated_request.body)
+
+      assert decoded["verbosity"] == "low"
+    end
+
+    test "encode_body includes verbosity when provided as string" do
+      {:ok, model} = ReqLLM.model("openai:gpt-4o")
+      context = context_fixture()
+
+      mock_request = %Req.Request{
+        options: [
+          context: context,
+          model: model.model,
+          provider_options: [verbosity: "high"]
+        ]
+      }
+
+      updated_request = OpenAI.encode_body(mock_request)
+      decoded = Jason.decode!(updated_request.body)
+
+      assert decoded["verbosity"] == "high"
+    end
+
+    test "encode_body omits verbosity when not provided" do
+      {:ok, model} = ReqLLM.model("openai:gpt-4o")
+      context = context_fixture()
+
+      mock_request = %Req.Request{
+        options: [
+          context: context,
+          model: model.model
+        ]
+      }
+
+      updated_request = OpenAI.encode_body(mock_request)
+      decoded = Jason.decode!(updated_request.body)
+
+      refute Map.has_key?(decoded, "verbosity")
+    end
   end
 
   describe "response decoding" do