Add support for Verbosity introduced in GPT-5

src/AI.Tests/OpenAITests.cs

@@ -156,7 +156,7 @@ public async Task GPT5_ThinkingTime(ReasoningEffort effort)
         Assert.StartsWith("gpt-5", response.ModelId);
         Assert.DoesNotContain("nano", response.ModelId);
 
-        // Reasoning should have been set to medium
+        // Reasoning should have been set to expected value
         Assert.All(requests, x =>
         {
             var search = Assert.IsType<JsonObject>(x["reasoning"]);
@@ -166,6 +166,48 @@ public async Task GPT5_ThinkingTime(ReasoningEffort effort)
         output.WriteLine($"Effort: {effort}, Time: {watch.ElapsedMilliseconds}ms, Tokens: {response.Usage?.TotalTokenCount}");
     }
 
+    [SecretsTheory("OPENAI_API_KEY")]
+    [InlineData(Verbosity.Low)]
+    [InlineData(Verbosity.Medium)]
+    [InlineData(Verbosity.High)]
+    public async Task GPT5_Verbosity(Verbosity verbosity)
+    {
+        var messages = new Chat()
+        {
+            { "system", "You are an intelligent AI assistant that's an expert on everything." },
+            { "user", "What's the answer to the universe and everything?" },
+        };
+
+        var requests = new List<JsonNode>();
+
+        var chat = new OpenAIChatClient(Configuration["OPENAI_API_KEY"]!, "gpt-5-nano",
+            OpenAIClientOptions.Observable(requests.Add).WriteTo(output));
+
+        var options = new ChatOptions
+        {
+            ModelId = "gpt-5-mini",
+            Verbosity = verbosity
+        };
+
+        var watch = System.Diagnostics.Stopwatch.StartNew();
+        var response = await chat.GetResponseAsync(messages, options);
+        watch.Stop();
+
+        var text = response.Text;
+        output.WriteLine(text);
+
+        Assert.StartsWith("gpt-5", response.ModelId);
+        Assert.DoesNotContain("nano", response.ModelId);
+
+        // Verbosity should have been set to the expected value
+        Assert.All(requests, x =>
+        {
+            var text = Assert.IsType<JsonObject>(x["text"]);
+            Assert.Equal(verbosity.ToString().ToLowerInvariant(), text["verbosity"]?.GetValue<string>());
+        });
+
+        output.WriteLine($"Verbosity: {verbosity}, Time: {watch.ElapsedMilliseconds}ms, Tokens: {response.Usage?.TotalTokenCount}");
+    }
 
     [SecretsFact("OPENAI_API_KEY")]
     public async Task WebSearchCountryHighContext()

src/AI/ChatExtensions.cs

@@ -43,5 +43,25 @@ public ReasoningEffort? ReasoningEffort
                 }
             }
         }
+
+        /// <summary>
+        /// Sets the <see cref="Verbosity"/> level for a GPT-5 model when generating responses, if supported
+        /// </summary>
+        public Verbosity? Verbosity
+        {
+            get => options.AdditionalProperties?.TryGetValue("verbosity", out var value) == true && value is Verbosity verbosity ? verbosity : null;
+            set
+            {
+                if (value is not null)
+                {
+                    options.AdditionalProperties ??= [];
+                    options.AdditionalProperties["verbosity"] = value;
+                }
+                else
+                {
+                    options.AdditionalProperties?.Remove("verbosity");
+                }
+            }
+        }
     }
 }

src/AI/OpenAI/OpenAIChatClient.cs

@@ -47,22 +47,18 @@ IChatClient GetChatClient(string modelId) => clients.GetOrAdd(modelId, model
         if (options is null)
             return null;
 
-        if (options.ReasoningEffort is ReasoningEffort effort)
+        if (options.ReasoningEffort.HasValue || options.Verbosity.HasValue)
         {
-            options.RawRepresentationFactory = _ => new ResponseCreationOptions
+            options.RawRepresentationFactory = _ =>
             {
-                ReasoningOptions = new ReasoningEffortOptions(effort)
-                //ReasoningOptions = new ResponseReasoningOptions()
-                //{
-                //    ReasoningEffortLevel = effort switch
-                //    {
-                //        ReasoningEffort.High => ResponseReasoningEffortLevel.High,
-                //        ReasoningEffort.Medium => ResponseReasoningEffortLevel.Medium,
-                //        // TODO: not exposed yet in the OpenAI package
-                //        // ReasoningEffort.Minimal => ResponseReasoningEffortLevel.Minimal,
-                //        _ => ResponseReasoningEffortLevel.Low
-                //    },
-                //}
+                var creation = new ResponseCreationOptions();
+                if (options.ReasoningEffort.HasValue)
+                    creation.ReasoningOptions = new ReasoningEffortOptions(options.ReasoningEffort!.Value);
+
+                if (options.Verbosity.HasValue)
+                    creation.TextOptions = new VerbosityOptions(options.Verbosity!.Value);
+
+                return creation;
             };
         }
 
@@ -85,4 +81,14 @@ protected override void JsonModelWriteCore(Utf8JsonWriter writer, ModelReaderWri
             base.JsonModelWriteCore(writer, options);
         }
     }
+
+    class VerbosityOptions(Verbosity verbosity) : ResponseTextOptions
+    {
+        protected override void JsonModelWriteCore(Utf8JsonWriter writer, ModelReaderWriterOptions options)
+        {
+            writer.WritePropertyName("verbosity"u8);
+            writer.WriteStringValue(verbosity.ToString().ToLowerInvariant());
+            base.JsonModelWriteCore(writer, options);
+        }
+    }
 }

src/AI/Verbosity.cs

@@ -0,0 +1,13 @@
+namespace Devlooped.Extensions.AI;
+
+/// <summary>
+/// Verbosity determines how many output tokens are generated for models that support it, such as GPT-5.
+/// Lowering the number of tokens reduces overall latency.
+/// </summary>
+/// <see href="https://platform.openai.com/docs/guides/latest-model?utm_source=chatgpt.com#verbosity"/>
+public enum Verbosity
+{
+    Low,
+    Medium,
+    High
+}