Add service tier support to Chat and Responses APIs

api/OpenAI.net8.0.cs

@@ -1511,6 +1511,8 @@ public class ChatCompletion : IJsonModel<ChatCompletion>, IPersistableModel<Chat
         public string Refusal { get; }
         public IReadOnlyList<ChatTokenLogProbabilityDetails> RefusalTokenLogProbabilities { get; }
         public ChatMessageRole Role { get; }
+        [Experimental("OPENAI001")]
+        public ChatServiceTier? ServiceTier { get; }
         public string SystemFingerprint { get; }
         public IReadOnlyList<ChatToolCall> ToolCalls { get; }
         public ChatTokenUsage Usage { get; }
@@ -1626,6 +1628,8 @@ public class ChatCompletionOptions : IJsonModel<ChatCompletionOptions>, IPersist
         public ChatResponseModalities ResponseModalities { get; set; }
         [Experimental("OPENAI001")]
         public long? Seed { get; set; }
+        [Experimental("OPENAI001")]
+        public ChatServiceTier? ServiceTier { get; set; }
         public IList<string> StopSequences { get; }
         public bool? StoredOutputEnabled { get; set; }
         public float? Temperature { get; set; }
@@ -1974,6 +1978,24 @@ public enum ChatResponseModalities {
         Text = 1,
         Audio = 2
     }
+    [Experimental("OPENAI001")]
+    public readonly partial struct ChatServiceTier : IEquatable<ChatServiceTier> {
+        public ChatServiceTier(string value);
+        public static ChatServiceTier Auto { get; }
+        public static ChatServiceTier Default { get; }
+        public static ChatServiceTier Flex { get; }
+        public static ChatServiceTier Scale { get; }
+        public readonly bool Equals(ChatServiceTier other);
+        [EditorBrowsable(EditorBrowsableState.Never)]
+        public override readonly bool Equals(object obj);
+        [EditorBrowsable(EditorBrowsableState.Never)]
+        public override readonly int GetHashCode();
+        public static bool operator ==(ChatServiceTier left, ChatServiceTier right);
+        public static implicit operator ChatServiceTier(string value);
+        public static implicit operator ChatServiceTier?(string value);
+        public static bool operator !=(ChatServiceTier left, ChatServiceTier right);
+        public override readonly string ToString();
+    }
     public class ChatTokenLogProbabilityDetails : IJsonModel<ChatTokenLogProbabilityDetails>, IPersistableModel<ChatTokenLogProbabilityDetails> {
         public float LogProbability { get; }
         public string Token { get; }
@@ -2143,6 +2165,8 @@ public class StreamingChatCompletionUpdate : IJsonModel<StreamingChatCompletionU
         public IReadOnlyList<ChatTokenLogProbabilityDetails> RefusalTokenLogProbabilities { get; }
         public string RefusalUpdate { get; }
         public ChatMessageRole? Role { get; }
+        [Experimental("OPENAI001")]
+        public ChatServiceTier? ServiceTier { get; }
         public string SystemFingerprint { get; }
         public IReadOnlyList<StreamingChatToolCallUpdate> ToolCallUpdates { get; }
         public ChatTokenUsage Usage { get; }
@@ -4772,6 +4796,7 @@ public class OpenAIResponse : IJsonModel<OpenAIResponse>, IPersistableModel<Open
         public bool ParallelToolCallsEnabled { get; }
         public string PreviousResponseId { get; }
         public ResponseReasoningOptions ReasoningOptions { get; }
+        public ResponseServiceTier? ServiceTier { get; }
         public ResponseStatus? Status { get; }
         public float? Temperature { get; }
         public ResponseTextOptions TextOptions { get; }
@@ -4926,6 +4951,7 @@ public class ResponseCreationOptions : IJsonModel<ResponseCreationOptions>, IPer
         public bool? ParallelToolCallsEnabled { get; set; }
         public string PreviousResponseId { get; set; }
         public ResponseReasoningOptions ReasoningOptions { get; set; }
+        public ResponseServiceTier? ServiceTier { get; set; }
         public bool? StoredOutputEnabled { get; set; }
         public float? Temperature { get; set; }
         public ResponseTextOptions TextOptions { get; set; }
@@ -5171,6 +5197,24 @@ public class ResponseReasoningOptions : IJsonModel<ResponseReasoningOptions>, IP
         public override readonly string ToString();
     }
     [Experimental("OPENAI001")]
+    public readonly partial struct ResponseServiceTier : IEquatable<ResponseServiceTier> {
+        public ResponseServiceTier(string value);
+        public static ResponseServiceTier Auto { get; }
+        public static ResponseServiceTier Default { get; }
+        public static ResponseServiceTier Flex { get; }
+        public static ResponseServiceTier Scale { get; }
+        public readonly bool Equals(ResponseServiceTier other);
+        [EditorBrowsable(EditorBrowsableState.Never)]
+        public override readonly bool Equals(object obj);
+        [EditorBrowsable(EditorBrowsableState.Never)]
+        public override readonly int GetHashCode();
+        public static bool operator ==(ResponseServiceTier left, ResponseServiceTier right);
+        public static implicit operator ResponseServiceTier(string value);
+        public static implicit operator ResponseServiceTier?(string value);
+        public static bool operator !=(ResponseServiceTier left, ResponseServiceTier right);
+        public override readonly string ToString();
+    }
+    [Experimental("OPENAI001")]
     public enum ResponseStatus {
         InProgress = 0,
         Completed = 1,

api/OpenAI.netstandard2.0.cs

@@ -1345,6 +1345,7 @@ public class ChatCompletion : IJsonModel<ChatCompletion>, IPersistableModel<Chat
         public string Refusal { get; }
         public IReadOnlyList<ChatTokenLogProbabilityDetails> RefusalTokenLogProbabilities { get; }
         public ChatMessageRole Role { get; }
+        public ChatServiceTier? ServiceTier { get; }
         public string SystemFingerprint { get; }
         public IReadOnlyList<ChatToolCall> ToolCalls { get; }
         public ChatTokenUsage Usage { get; }
@@ -1444,6 +1445,7 @@ public class ChatCompletionOptions : IJsonModel<ChatCompletionOptions>, IPersist
         public ChatResponseFormat ResponseFormat { get; set; }
         public ChatResponseModalities ResponseModalities { get; set; }
         public long? Seed { get; set; }
+        public ChatServiceTier? ServiceTier { get; set; }
         public IList<string> StopSequences { get; }
         public bool? StoredOutputEnabled { get; set; }
         public float? Temperature { get; set; }
@@ -1731,6 +1733,23 @@ public enum ChatResponseModalities {
         Text = 1,
         Audio = 2
     }
+    public readonly partial struct ChatServiceTier : IEquatable<ChatServiceTier> {
+        public ChatServiceTier(string value);
+        public static ChatServiceTier Auto { get; }
+        public static ChatServiceTier Default { get; }
+        public static ChatServiceTier Flex { get; }
+        public static ChatServiceTier Scale { get; }
+        public readonly bool Equals(ChatServiceTier other);
+        [EditorBrowsable(EditorBrowsableState.Never)]
+        public override readonly bool Equals(object obj);
+        [EditorBrowsable(EditorBrowsableState.Never)]
+        public override readonly int GetHashCode();
+        public static bool operator ==(ChatServiceTier left, ChatServiceTier right);
+        public static implicit operator ChatServiceTier(string value);
+        public static implicit operator ChatServiceTier?(string value);
+        public static bool operator !=(ChatServiceTier left, ChatServiceTier right);
+        public override readonly string ToString();
+    }
     public class ChatTokenLogProbabilityDetails : IJsonModel<ChatTokenLogProbabilityDetails>, IPersistableModel<ChatTokenLogProbabilityDetails> {
         public float LogProbability { get; }
         public string Token { get; }
@@ -1863,6 +1882,7 @@ public class StreamingChatCompletionUpdate : IJsonModel<StreamingChatCompletionU
         public IReadOnlyList<ChatTokenLogProbabilityDetails> RefusalTokenLogProbabilities { get; }
         public string RefusalUpdate { get; }
         public ChatMessageRole? Role { get; }
+        public ChatServiceTier? ServiceTier { get; }
         public string SystemFingerprint { get; }
         public IReadOnlyList<StreamingChatToolCallUpdate> ToolCallUpdates { get; }
         public ChatTokenUsage Usage { get; }
@@ -4207,6 +4227,7 @@ public class OpenAIResponse : IJsonModel<OpenAIResponse>, IPersistableModel<Open
         public bool ParallelToolCallsEnabled { get; }
         public string PreviousResponseId { get; }
         public ResponseReasoningOptions ReasoningOptions { get; }
+        public ResponseServiceTier? ServiceTier { get; }
         public ResponseStatus? Status { get; }
         public float? Temperature { get; }
         public ResponseTextOptions TextOptions { get; }
@@ -4351,6 +4372,7 @@ public class ResponseCreationOptions : IJsonModel<ResponseCreationOptions>, IPer
         public bool? ParallelToolCallsEnabled { get; set; }
         public string PreviousResponseId { get; set; }
         public ResponseReasoningOptions ReasoningOptions { get; set; }
+        public ResponseServiceTier? ServiceTier { get; set; }
         public bool? StoredOutputEnabled { get; set; }
         public float? Temperature { get; set; }
         public ResponseTextOptions TextOptions { get; set; }
@@ -4575,6 +4597,23 @@ public class ResponseReasoningOptions : IJsonModel<ResponseReasoningOptions>, IP
         public static bool operator !=(ResponseReasoningSummaryVerbosity left, ResponseReasoningSummaryVerbosity right);
         public override readonly string ToString();
     }
+    public readonly partial struct ResponseServiceTier : IEquatable<ResponseServiceTier> {
+        public ResponseServiceTier(string value);
+        public static ResponseServiceTier Auto { get; }
+        public static ResponseServiceTier Default { get; }
+        public static ResponseServiceTier Flex { get; }
+        public static ResponseServiceTier Scale { get; }
+        public readonly bool Equals(ResponseServiceTier other);
+        [EditorBrowsable(EditorBrowsableState.Never)]
+        public override readonly bool Equals(object obj);
+        [EditorBrowsable(EditorBrowsableState.Never)]
+        public override readonly int GetHashCode();
+        public static bool operator ==(ResponseServiceTier left, ResponseServiceTier right);
+        public static implicit operator ResponseServiceTier(string value);
+        public static implicit operator ResponseServiceTier?(string value);
+        public static bool operator !=(ResponseServiceTier left, ResponseServiceTier right);
+        public override readonly string ToString();
+    }
     public enum ResponseStatus {
         InProgress = 0,
         Completed = 1,

specification/client/chat.client.tsp

@@ -18,3 +18,7 @@ using Azure.ClientGenerator.Core;
 @@visibility(ChatCompletionStreamResponseDelta.tool_calls, Lifecycle.Read);
 
 @@clientName(Chat.createChatCompletion, "CompleteChat");
+
+@@alternateType(CreateChatCompletionRequest.service_tier, DotNetChatServiceTier);
+@@alternateType(CreateChatCompletionResponse.service_tier, DotNetChatServiceTier);
+@@alternateType(CreateChatCompletionStreamResponse.service_tier, DotNetChatServiceTier);

specification/client/models/common.models.tsp

@@ -153,4 +153,12 @@ alias CollectionBeforeQueryParameter = {
    * subsequent call can include before=obj_foo in order to fetch the previous page of the list.
    */
   @query beforeId?: string;
-};
+};
+
+union DotNetChatServiceTier {
+   ServiceTier
+}
+
+union DotNetResponseServiceTier {
+  ServiceTier
+}

specification/client/responses.client.tsp

@@ -0,0 +1,8 @@
+import "../base/typespec/responses/main.tsp";
+import "@azure-tools/typespec-client-generator-core";
+
+using OpenAI;
+using Azure.ClientGenerator.Core;
+
+@@alternateType(CreateResponse.service_tier, DotNetResponseServiceTier);
+@@alternateType(Response.service_tier, DotNetResponseServiceTier);

specification/main.tsp

@@ -10,6 +10,7 @@ import "./client/graders.client.tsp";
 import "./client/images.client.tsp";
 import "./client/messages.client.tsp";
 import "./client/moderations.client.tsp";
+import "./client/responses.client.tsp";
 import "./client/runs.client.tsp";
 import "./client/threads.client.tsp";
 import "./client/vector-stores.client.tsp";

src/Custom/Chat/ChatCompletion.cs

@@ -21,10 +21,6 @@ public partial class ChatCompletion
     [CodeGenMember("Object")]
     internal string Object { get; } = "chat.completion";
 
-    // CUSTOM: Made internal.
-    [CodeGenMember("ServiceTier")]
-    internal InternalServiceTier? ServiceTier { get; }
-
     // CUSTOM: Made internal. We only get back a single choice, and instead we flatten the structure for usability.
     [CodeGenMember("Choices")]
     internal IReadOnlyList<InternalCreateChatCompletionResponseChoice> Choices { get; }

src/Custom/Chat/ChatCompletionOptions.cs

@@ -107,9 +107,6 @@ public partial class ChatCompletionOptions
     [CodeGenMember("ParallelToolCalls")]
     public bool? AllowParallelToolCalls { get; set; }
 
-    [CodeGenMember("ServiceTier")]
-    internal InternalServiceTier? _serviceTier;
-
     // CUSTOM: Renamed.
     /// <summary>
     ///     A unique identifier representing your end-user, which can help OpenAI to monitor and detect abuse.

src/Custom/Chat/ChatServiceTier.cs

@@ -0,0 +1,7 @@
+namespace OpenAI.Chat;
+
+// CUSTOM: Renamed.
+[CodeGenType("DotNetChatServiceTier")]
+public readonly partial struct ChatServiceTier
+{
+}

src/Custom/Chat/OpenAIChatModelFactory.cs

@@ -96,10 +96,10 @@ public static ChatCompletion ChatCompletion(
         return new ChatCompletion(
             id,
             model,
+            serviceTier: null,
             systemFingerprint,
             usage,
             "chat.completion",
-            serviceTier: null,
             choices,
             createdAt,
             additionalBinaryDataProperties: null);
@@ -311,10 +311,10 @@ public static StreamingChatCompletionUpdate StreamingChatCompletionUpdate(
 
         return new StreamingChatCompletionUpdate(
             model,
+            serviceTier: null,
             systemFingerprint,
             "chat.completion.chunk",
             completionId,
-            serviceTier: null,
             choices,
             createdAt,
             usage,

src/Custom/Chat/Streaming/StreamingChatCompletionUpdate.cs

@@ -29,10 +29,6 @@ public partial class StreamingChatCompletionUpdate
     [CodeGenMember("Id")]
     public string CompletionId { get; }
 
-    // CUSTOM: Made internal.
-    [CodeGenMember("ServiceTier")]
-    internal InternalServiceTier? ServiceTier { get; }
-
     // CUSTOM: Made internal.We only get back a single choice, and instead we flatten the structure for usability.
     [CodeGenMember("Choices")]
     internal IReadOnlyList<InternalCreateChatCompletionStreamResponseChoice> Choices { get; }

src/Custom/Common/Internal/GeneratorStubs.cs

@@ -11,5 +11,4 @@ namespace OpenAI.Internal;
 [CodeGenType("WebSearchContextSize")] internal readonly partial struct InternalWebSearchContextSize { }
 [CodeGenType("LogProbProperties")] internal partial class InternalLogProbProperties { }
 [CodeGenType("ModelIdsShared")] internal readonly partial struct InternalModelIdsShared { }
-[CodeGenType("VoiceIdsShared")] internal readonly partial struct InternalVoiceIdsShared { }
-[CodeGenType("ServiceTier")] internal readonly partial struct InternalServiceTier { }
+[CodeGenType("VoiceIdsShared")] internal readonly partial struct InternalVoiceIdsShared { }

src/Custom/Responses/ResponseServiceTier.cs

@@ -0,0 +1,7 @@
+namespace OpenAI.Responses;
+
+// CUSTOM: Renamed.
+[CodeGenType("DotNetResponseServiceTier")]
+public readonly partial struct ResponseServiceTier
+{
+}

src/Generated/Models/Chat/ChatCompletion.Serialization.cs

@@ -9,13 +9,12 @@
 using System.Diagnostics.CodeAnalysis;
 using System.Text.Json;
 using OpenAI;
-using OpenAI.Internal;
 
 namespace OpenAI.Chat
 {
     public partial class ChatCompletion : IJsonModel<ChatCompletion>
     {
-        internal ChatCompletion() : this(null, null, null, null, null, default, null, default, null)
+        internal ChatCompletion() : this(null, null, default, null, null, null, null, default, null)
         {
         }
 
@@ -44,6 +43,11 @@ protected virtual void JsonModelWriteCore(Utf8JsonWriter writer, ModelReaderWrit
                 writer.WritePropertyName("model"u8);
                 writer.WriteStringValue(Model);
             }
+            if (Optional.IsDefined(ServiceTier) && _additionalBinaryDataProperties?.ContainsKey("service_tier") != true)
+            {
+                writer.WritePropertyName("service_tier"u8);
+                writer.WriteStringValue(ServiceTier.Value.ToString());
+            }
             if (Optional.IsDefined(SystemFingerprint) && _additionalBinaryDataProperties?.ContainsKey("system_fingerprint") != true)
             {
                 writer.WritePropertyName("system_fingerprint"u8);
@@ -59,11 +63,6 @@ protected virtual void JsonModelWriteCore(Utf8JsonWriter writer, ModelReaderWrit
                 writer.WritePropertyName("object"u8);
                 writer.WriteStringValue(Object);
             }
-            if (Optional.IsDefined(ServiceTier) && _additionalBinaryDataProperties?.ContainsKey("service_tier") != true)
-            {
-                writer.WritePropertyName("service_tier"u8);
-                writer.WriteStringValue(ServiceTier.Value.ToString());
-            }
             if (_additionalBinaryDataProperties?.ContainsKey("choices") != true)
             {
                 writer.WritePropertyName("choices"u8);
@@ -123,10 +122,10 @@ internal static ChatCompletion DeserializeChatCompletion(JsonElement element, Mo
             }
             string id = default;
             string model = default;
+            ChatServiceTier? serviceTier = default;
             string systemFingerprint = default;
             ChatTokenUsage usage = default;
             string @object = default;
-            InternalServiceTier? serviceTier = default;
             IReadOnlyList<InternalCreateChatCompletionResponseChoice> choices = default;
             DateTimeOffset createdAt = default;
             IDictionary<string, BinaryData> additionalBinaryDataProperties = new ChangeTrackingDictionary<string, BinaryData>();
@@ -142,6 +141,15 @@ internal static ChatCompletion DeserializeChatCompletion(JsonElement element, Mo
                     model = prop.Value.GetString();
                     continue;
                 }
+                if (prop.NameEquals("service_tier"u8))
+                {
+                    if (prop.Value.ValueKind == JsonValueKind.Null)
+                    {
+                        continue;
+                    }
+                    serviceTier = new ChatServiceTier(prop.Value.GetString());
+                    continue;
+                }
                 if (prop.NameEquals("system_fingerprint"u8))
                 {
                     systemFingerprint = prop.Value.GetString();
@@ -161,16 +169,6 @@ internal static ChatCompletion DeserializeChatCompletion(JsonElement element, Mo
                     @object = prop.Value.GetString();
                     continue;
                 }
-                if (prop.NameEquals("service_tier"u8))
-                {
-                    if (prop.Value.ValueKind == JsonValueKind.Null)
-                    {
-                        serviceTier = null;
-                        continue;
-                    }
-                    serviceTier = new InternalServiceTier(prop.Value.GetString());
-                    continue;
-                }
                 if (prop.NameEquals("choices"u8))
                 {
                     List<InternalCreateChatCompletionResponseChoice> array = new List<InternalCreateChatCompletionResponseChoice>();
@@ -192,10 +190,10 @@ internal static ChatCompletion DeserializeChatCompletion(JsonElement element, Mo
             return new ChatCompletion(
                 id,
                 model,
+                serviceTier,
                 systemFingerprint,
                 usage,
                 @object,
-                serviceTier,
                 choices,
                 createdAt,
                 additionalBinaryDataProperties);