openai 0.60.0 → 0.62.0

data/rbi/openai/models/realtime/audio_transcription.rbi

@@ -12,6 +12,23 @@ module OpenAI
             )
           end
 
+        # Controls how long the model waits before emitting transcription text. Higher
+        # values can improve transcription accuracy at the cost of latency. Only supported
+        # with `gpt-realtime-whisper` in GA Realtime sessions.
+        sig do
+          returns(
+            T.nilable(OpenAI::Realtime::AudioTranscription::Delay::OrSymbol)
+          )
+        end
+        attr_reader :delay
+
+        sig do
+          params(
+            delay: OpenAI::Realtime::AudioTranscription::Delay::OrSymbol
+          ).void
+        end
+        attr_writer :delay
+
         # The language of the input audio. Supplying the input language in
         # [ISO-639-1](https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) (e.g. `en`)
         # format will improve accuracy and latency.
@@ -23,8 +40,8 @@ module OpenAI
 
         # The model to use for transcription. Current options are `whisper-1`,
         # `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`,
-        # `gpt-4o-transcribe`, and `gpt-4o-transcribe-diarize`. Use
-        # `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.
+        # `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`.
+        # Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.
         sig do
           returns(
             T.nilable(
@@ -53,6 +70,7 @@ module OpenAI
         # [prompt is a list of keywords](https://platform.openai.com/docs/guides/speech-to-text#prompting).
         # For `gpt-4o-transcribe` models (excluding `gpt-4o-transcribe-diarize`), the
         # prompt is a free text string, for example "expect words related to technology".
+        # Prompt is not supported with `gpt-realtime-whisper` in GA Realtime sessions.
         sig { returns(T.nilable(String)) }
         attr_reader :prompt
 
@@ -61,6 +79,7 @@ module OpenAI
 
         sig do
           params(
+            delay: OpenAI::Realtime::AudioTranscription::Delay::OrSymbol,
             language: String,
             model:
               T.any(
@@ -71,20 +90,25 @@ module OpenAI
           ).returns(T.attached_class)
         end
         def self.new(
+          # Controls how long the model waits before emitting transcription text. Higher
+          # values can improve transcription accuracy at the cost of latency. Only supported
+          # with `gpt-realtime-whisper` in GA Realtime sessions.
+          delay: nil,
           # The language of the input audio. Supplying the input language in
           # [ISO-639-1](https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) (e.g. `en`)
           # format will improve accuracy and latency.
           language: nil,
           # The model to use for transcription. Current options are `whisper-1`,
           # `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`,
-          # `gpt-4o-transcribe`, and `gpt-4o-transcribe-diarize`. Use
-          # `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.
+          # `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`.
+          # Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.
           model: nil,
           # An optional text to guide the model's style or continue a previous audio
           # segment. For `whisper-1`, the
           # [prompt is a list of keywords](https://platform.openai.com/docs/guides/speech-to-text#prompting).
           # For `gpt-4o-transcribe` models (excluding `gpt-4o-transcribe-diarize`), the
           # prompt is a free text string, for example "expect words related to technology".
+          # Prompt is not supported with `gpt-realtime-whisper` in GA Realtime sessions.
           prompt: nil
         )
         end
@@ -92,6 +116,7 @@ module OpenAI
         sig do
           override.returns(
             {
+              delay: OpenAI::Realtime::AudioTranscription::Delay::OrSymbol,
               language: String,
               model:
                 T.any(
@@ -105,10 +130,59 @@ module OpenAI
         def to_hash
         end
 
+        # Controls how long the model waits before emitting transcription text. Higher
+        # values can improve transcription accuracy at the cost of latency. Only supported
+        # with `gpt-realtime-whisper` in GA Realtime sessions.
+        module Delay
+          extend OpenAI::Internal::Type::Enum
+
+          TaggedSymbol =
+            T.type_alias do
+              T.all(Symbol, OpenAI::Realtime::AudioTranscription::Delay)
+            end
+          OrSymbol = T.type_alias { T.any(Symbol, String) }
+
+          MINIMAL =
+            T.let(
+              :minimal,
+              OpenAI::Realtime::AudioTranscription::Delay::TaggedSymbol
+            )
+          LOW =
+            T.let(
+              :low,
+              OpenAI::Realtime::AudioTranscription::Delay::TaggedSymbol
+            )
+          MEDIUM =
+            T.let(
+              :medium,
+              OpenAI::Realtime::AudioTranscription::Delay::TaggedSymbol
+            )
+          HIGH =
+            T.let(
+              :high,
+              OpenAI::Realtime::AudioTranscription::Delay::TaggedSymbol
+            )
+          XHIGH =
+            T.let(
+              :xhigh,
+              OpenAI::Realtime::AudioTranscription::Delay::TaggedSymbol
+            )
+
+          sig do
+            override.returns(
+              T::Array[
+                OpenAI::Realtime::AudioTranscription::Delay::TaggedSymbol
+              ]
+            )
+          end
+          def self.values
+          end
+        end
+
         # The model to use for transcription. Current options are `whisper-1`,
         # `gpt-4o-mini-transcribe`, `gpt-4o-mini-transcribe-2025-12-15`,
-        # `gpt-4o-transcribe`, and `gpt-4o-transcribe-diarize`. Use
-        # `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.
+        # `gpt-4o-transcribe`, `gpt-4o-transcribe-diarize`, and `gpt-realtime-whisper`.
+        # Use `gpt-4o-transcribe-diarize` when you need diarization with speaker labels.
         module Model
           extend OpenAI::Internal::Type::Union
 
@@ -159,6 +233,11 @@ module OpenAI
               :"gpt-4o-transcribe-diarize",
               OpenAI::Realtime::AudioTranscription::Model::TaggedSymbol
             )
+          GPT_REALTIME_WHISPER =
+            T.let(
+              :"gpt-realtime-whisper",
+              OpenAI::Realtime::AudioTranscription::Model::TaggedSymbol
+            )
         end
       end
     end

data/rbi/openai/models/realtime/realtime_audio_config_input.rbi

@@ -91,6 +91,9 @@ module OpenAI
         # trails off with "uhhm", the model will score a low probability of turn end and
         # wait longer for the user to continue speaking. This can be useful for more
         # natural conversations, but may have a higher latency.
+        #
+        # For `gpt-realtime-whisper` transcription sessions, turn detection must be set to
+        # `null`; VAD is not supported.
         sig do
           returns(
             T.nilable(
@@ -154,6 +157,9 @@ module OpenAI
           # trails off with "uhhm", the model will score a low probability of turn end and
           # wait longer for the user to continue speaking. This can be useful for more
           # natural conversations, but may have a higher latency.
+          #
+          # For `gpt-realtime-whisper` transcription sessions, turn detection must be set to
+          # `null`; VAD is not supported.
           turn_detection: nil
         )
         end

data/rbi/openai/models/realtime/realtime_audio_input_turn_detection.rbi

@@ -16,6 +16,9 @@ module OpenAI
       # trails off with "uhhm", the model will score a low probability of turn end and
       # wait longer for the user to continue speaking. This can be useful for more
       # natural conversations, but may have a higher latency.
+      #
+      # For `gpt-realtime-whisper` transcription sessions, turn detection must be set to
+      # `null`; VAD is not supported.
       module RealtimeAudioInputTurnDetection
         extend OpenAI::Internal::Type::Union
 

data/rbi/openai/models/realtime/realtime_reasoning.rbi

@@ -0,0 +1,54 @@
+# typed: strong
+
+module OpenAI
+  module Models
+    module Realtime
+      class RealtimeReasoning < OpenAI::Internal::Type::BaseModel
+        OrHash =
+          T.type_alias do
+            T.any(
+              OpenAI::Realtime::RealtimeReasoning,
+              OpenAI::Internal::AnyHash
+            )
+          end
+
+        # Constrains effort on reasoning for reasoning-capable Realtime models such as
+        # `gpt-realtime-2`.
+        sig do
+          returns(
+            T.nilable(OpenAI::Realtime::RealtimeReasoningEffort::OrSymbol)
+          )
+        end
+        attr_reader :effort
+
+        sig do
+          params(
+            effort: OpenAI::Realtime::RealtimeReasoningEffort::OrSymbol
+          ).void
+        end
+        attr_writer :effort
+
+        # Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.
+        sig do
+          params(
+            effort: OpenAI::Realtime::RealtimeReasoningEffort::OrSymbol
+          ).returns(T.attached_class)
+        end
+        def self.new(
+          # Constrains effort on reasoning for reasoning-capable Realtime models such as
+          # `gpt-realtime-2`.
+          effort: nil
+        )
+        end
+
+        sig do
+          override.returns(
+            { effort: OpenAI::Realtime::RealtimeReasoningEffort::OrSymbol }
+          )
+        end
+        def to_hash
+        end
+      end
+    end
+  end
+end

data/rbi/openai/models/realtime/realtime_reasoning_effort.rbi

@@ -0,0 +1,44 @@
+# typed: strong
+
+module OpenAI
+  module Models
+    module Realtime
+      # Constrains effort on reasoning for reasoning-capable Realtime models such as
+      # `gpt-realtime-2`.
+      module RealtimeReasoningEffort
+        extend OpenAI::Internal::Type::Enum
+
+        TaggedSymbol =
+          T.type_alias do
+            T.all(Symbol, OpenAI::Realtime::RealtimeReasoningEffort)
+          end
+        OrSymbol = T.type_alias { T.any(Symbol, String) }
+
+        MINIMAL =
+          T.let(
+            :minimal,
+            OpenAI::Realtime::RealtimeReasoningEffort::TaggedSymbol
+          )
+        LOW =
+          T.let(:low, OpenAI::Realtime::RealtimeReasoningEffort::TaggedSymbol)
+        MEDIUM =
+          T.let(
+            :medium,
+            OpenAI::Realtime::RealtimeReasoningEffort::TaggedSymbol
+          )
+        HIGH =
+          T.let(:high, OpenAI::Realtime::RealtimeReasoningEffort::TaggedSymbol)
+        XHIGH =
+          T.let(:xhigh, OpenAI::Realtime::RealtimeReasoningEffort::TaggedSymbol)
+
+        sig do
+          override.returns(
+            T::Array[OpenAI::Realtime::RealtimeReasoningEffort::TaggedSymbol]
+          )
+        end
+        def self.values
+        end
+      end
+    end
+  end
+end

data/rbi/openai/models/realtime/realtime_response_create_params.rbi

@@ -159,6 +159,14 @@ module OpenAI
         end
         attr_writer :output_modalities
 
+        # Whether the model may call multiple tools in parallel. Only supported by
+        # reasoning Realtime models such as `gpt-realtime-2`.
+        sig { returns(T.nilable(T::Boolean)) }
+        attr_reader :parallel_tool_calls
+
+        sig { params(parallel_tool_calls: T::Boolean).void }
+        attr_writer :parallel_tool_calls
+
         # Reference to a prompt template and its variables.
         # [Learn more](https://platform.openai.com/docs/guides/text?api-mode=responses#reusable-prompts).
         sig { returns(T.nilable(OpenAI::Responses::ResponsePrompt)) }
@@ -171,6 +179,15 @@ module OpenAI
         end
         attr_writer :prompt
 
+        # Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.
+        sig { returns(T.nilable(OpenAI::Realtime::RealtimeReasoning)) }
+        attr_reader :reasoning
+
+        sig do
+          params(reasoning: OpenAI::Realtime::RealtimeReasoning::OrHash).void
+        end
+        attr_writer :reasoning
+
         # How the model chooses tools. Provide one of the string modes or force a specific
         # function/MCP tool.
         sig do
@@ -256,7 +273,9 @@ module OpenAI
               T::Array[
                 OpenAI::Realtime::RealtimeResponseCreateParams::OutputModality::OrSymbol
               ],
+            parallel_tool_calls: T::Boolean,
             prompt: T.nilable(OpenAI::Responses::ResponsePrompt::OrHash),
+            reasoning: OpenAI::Realtime::RealtimeReasoning::OrHash,
             tool_choice:
               T.any(
                 OpenAI::Responses::ToolChoiceOptions::OrSymbol,
@@ -313,9 +332,14 @@ module OpenAI
           # transcript. Setting the output to mode `text` will disable audio output from the
           # model.
           output_modalities: nil,
+          # Whether the model may call multiple tools in parallel. Only supported by
+          # reasoning Realtime models such as `gpt-realtime-2`.
+          parallel_tool_calls: nil,
           # Reference to a prompt template and its variables.
           # [Learn more](https://platform.openai.com/docs/guides/text?api-mode=responses#reusable-prompts).
           prompt: nil,
+          # Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.
+          reasoning: nil,
           # How the model chooses tools. Provide one of the string modes or force a specific
           # function/MCP tool.
           tool_choice: nil,
@@ -354,7 +378,9 @@ module OpenAI
                 T::Array[
                   OpenAI::Realtime::RealtimeResponseCreateParams::OutputModality::OrSymbol
                 ],
+              parallel_tool_calls: T::Boolean,
               prompt: T.nilable(OpenAI::Responses::ResponsePrompt),
+              reasoning: OpenAI::Realtime::RealtimeReasoning,
               tool_choice:
                 T.any(
                   OpenAI::Responses::ToolChoiceOptions::OrSymbol,

data/rbi/openai/models/realtime/realtime_session.rbi

@@ -274,6 +274,9 @@ module OpenAI
         # trails off with "uhhm", the model will score a low probability of turn end and
         # wait longer for the user to continue speaking. This can be useful for more
         # natural conversations, but may have a higher latency.
+        #
+        # For `gpt-realtime-whisper` transcription sessions, turn detection must be set to
+        # `null`; VAD is not supported.
         sig do
           returns(
             T.nilable(
@@ -442,6 +445,9 @@ module OpenAI
           # trails off with "uhhm", the model will score a low probability of turn end and
           # wait longer for the user to continue speaking. This can be useful for more
           # natural conversations, but may have a higher latency.
+          #
+          # For `gpt-realtime-whisper` transcription sessions, turn detection must be set to
+          # `null`; VAD is not supported.
           turn_detection: nil,
           # The voice the model uses to respond. Voice cannot be changed during the session
           # once the model has responded with audio at least once. Current voice options are
@@ -948,6 +954,9 @@ module OpenAI
         # trails off with "uhhm", the model will score a low probability of turn end and
         # wait longer for the user to continue speaking. This can be useful for more
         # natural conversations, but may have a higher latency.
+        #
+        # For `gpt-realtime-whisper` transcription sessions, turn detection must be set to
+        # `null`; VAD is not supported.
         module TurnDetection
           extend OpenAI::Internal::Type::Union
 

data/rbi/openai/models/realtime/realtime_session_create_request.rbi

@@ -125,6 +125,14 @@ module OpenAI
         end
         attr_writer :output_modalities
 
+        # Whether the model may call multiple tools in parallel. Only supported by
+        # reasoning Realtime models such as `gpt-realtime-2`.
+        sig { returns(T.nilable(T::Boolean)) }
+        attr_reader :parallel_tool_calls
+
+        sig { params(parallel_tool_calls: T::Boolean).void }
+        attr_writer :parallel_tool_calls
+
         # Reference to a prompt template and its variables.
         # [Learn more](https://platform.openai.com/docs/guides/text?api-mode=responses#reusable-prompts).
         sig { returns(T.nilable(OpenAI::Responses::ResponsePrompt)) }
@@ -137,6 +145,15 @@ module OpenAI
         end
         attr_writer :prompt
 
+        # Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.
+        sig { returns(T.nilable(OpenAI::Realtime::RealtimeReasoning)) }
+        attr_reader :reasoning
+
+        sig do
+          params(reasoning: OpenAI::Realtime::RealtimeReasoning::OrHash).void
+        end
+        attr_writer :reasoning
+
         # How the model chooses tools. Provide one of the string modes or force a specific
         # function/MCP tool.
         sig do
@@ -271,7 +288,9 @@ module OpenAI
               T::Array[
                 OpenAI::Realtime::RealtimeSessionCreateRequest::OutputModality::OrSymbol
               ],
+            parallel_tool_calls: T::Boolean,
             prompt: T.nilable(OpenAI::Responses::ResponsePrompt::OrHash),
+            reasoning: OpenAI::Realtime::RealtimeReasoning::OrHash,
             tool_choice:
               T.any(
                 OpenAI::Responses::ToolChoiceOptions::OrSymbol,
@@ -331,9 +350,14 @@ module OpenAI
           # can be used to make the model respond with text only. It is not possible to
           # request both `text` and `audio` at the same time.
           output_modalities: nil,
+          # Whether the model may call multiple tools in parallel. Only supported by
+          # reasoning Realtime models such as `gpt-realtime-2`.
+          parallel_tool_calls: nil,
           # Reference to a prompt template and its variables.
           # [Learn more](https://platform.openai.com/docs/guides/text?api-mode=responses#reusable-prompts).
           prompt: nil,
+          # Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.
+          reasoning: nil,
           # How the model chooses tools. Provide one of the string modes or force a specific
           # function/MCP tool.
           tool_choice: nil,
@@ -391,7 +415,9 @@ module OpenAI
                 T::Array[
                   OpenAI::Realtime::RealtimeSessionCreateRequest::OutputModality::OrSymbol
                 ],
+              parallel_tool_calls: T::Boolean,
               prompt: T.nilable(OpenAI::Responses::ResponsePrompt),
+              reasoning: OpenAI::Realtime::RealtimeReasoning,
               tool_choice:
                 T.any(
                   OpenAI::Responses::ToolChoiceOptions::OrSymbol,
@@ -512,6 +538,11 @@ module OpenAI
               :"gpt-realtime-1.5",
               OpenAI::Realtime::RealtimeSessionCreateRequest::Model::TaggedSymbol
             )
+          GPT_REALTIME_2 =
+            T.let(
+              :"gpt-realtime-2",
+              OpenAI::Realtime::RealtimeSessionCreateRequest::Model::TaggedSymbol
+            )
           GPT_REALTIME_2025_08_28 =
             T.let(
               :"gpt-realtime-2025-08-28",

data/rbi/openai/models/realtime/realtime_session_create_response.rbi

@@ -14,16 +14,13 @@ module OpenAI
             )
           end
 
-        # Ephemeral key returned by the API.
-        sig { returns(OpenAI::Realtime::RealtimeSessionClientSecret) }
-        attr_reader :client_secret
+        # Unique identifier for the session that looks like `sess_1234567890abcdef`.
+        sig { returns(String) }
+        attr_accessor :id
 
-        sig do
-          params(
-            client_secret: OpenAI::Realtime::RealtimeSessionClientSecret::OrHash
-          ).void
-        end
-        attr_writer :client_secret
+        # The object type. Always `realtime.session`.
+        sig { returns(Symbol) }
+        attr_accessor :object
 
         # The type of session to create. Always `realtime` for the Realtime API.
         sig { returns(Symbol) }
@@ -45,6 +42,13 @@ module OpenAI
         end
         attr_writer :audio
 
+        # Expiration timestamp for the session, in seconds since epoch.
+        sig { returns(T.nilable(Integer)) }
+        attr_reader :expires_at
+
+        sig { params(expires_at: Integer).void }
+        attr_writer :expires_at
+
         # Additional fields to include in server outputs.
         #
         # `item.input_audio_transcription.logprobs`: Include logprobs for input audio
@@ -160,6 +164,15 @@ module OpenAI
         end
         attr_writer :prompt
 
+        # Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.
+        sig { returns(T.nilable(OpenAI::Realtime::RealtimeReasoning)) }
+        attr_reader :reasoning
+
+        sig do
+          params(reasoning: OpenAI::Realtime::RealtimeReasoning::OrHash).void
+        end
+        attr_writer :reasoning
+
         # How the model chooses tools. Provide one of the string modes or force a specific
         # function/MCP tool.
         sig do
@@ -258,14 +271,13 @@ module OpenAI
         end
         attr_writer :truncation
 
-        # A new Realtime session configuration, with an ephemeral key. Default TTL for
-        # keys is one minute.
+        # A Realtime session configuration object.
         sig do
           params(
-            client_secret:
-              OpenAI::Realtime::RealtimeSessionClientSecret::OrHash,
+            id: String,
             audio:
               OpenAI::Realtime::RealtimeSessionCreateResponse::Audio::OrHash,
+            expires_at: Integer,
             include:
               T::Array[
                 OpenAI::Realtime::RealtimeSessionCreateResponse::Include::OrSymbol
@@ -282,6 +294,7 @@ module OpenAI
                 OpenAI::Realtime::RealtimeSessionCreateResponse::OutputModality::OrSymbol
               ],
             prompt: T.nilable(OpenAI::Responses::ResponsePrompt::OrHash),
+            reasoning: OpenAI::Realtime::RealtimeReasoning::OrHash,
             tool_choice:
               T.any(
                 OpenAI::Responses::ToolChoiceOptions::OrSymbol,
@@ -307,14 +320,17 @@ module OpenAI
                 OpenAI::Realtime::RealtimeTruncation::RealtimeTruncationStrategy::OrSymbol,
                 OpenAI::Realtime::RealtimeTruncationRetentionRatio::OrHash
               ),
+            object: Symbol,
             type: Symbol
           ).returns(T.attached_class)
         end
         def self.new(
-          # Ephemeral key returned by the API.
-          client_secret:,
+          # Unique identifier for the session that looks like `sess_1234567890abcdef`.
+          id:,
           # Configuration for input and output audio.
           audio: nil,
+          # Expiration timestamp for the session, in seconds since epoch.
+          expires_at: nil,
           # Additional fields to include in server outputs.
           #
           # `item.input_audio_transcription.logprobs`: Include logprobs for input audio
@@ -346,6 +362,8 @@ module OpenAI
           # Reference to a prompt template and its variables.
           # [Learn more](https://platform.openai.com/docs/guides/text?api-mode=responses#reusable-prompts).
           prompt: nil,
+          # Configuration for reasoning-capable Realtime models such as `gpt-realtime-2`.
+          reasoning: nil,
           # How the model chooses tools. Provide one of the string modes or force a specific
           # function/MCP tool.
           tool_choice: nil,
@@ -378,6 +396,8 @@ module OpenAI
           # but would instead return an error if the conversation exceeds the model's input
           # token limit.
           truncation: nil,
+          # The object type. Always `realtime.session`.
+          object: :"realtime.session",
           # The type of session to create. Always `realtime` for the Realtime API.
           type: :realtime
         )
@@ -386,9 +406,11 @@ module OpenAI
         sig do
           override.returns(
             {
-              client_secret: OpenAI::Realtime::RealtimeSessionClientSecret,
+              id: String,
+              object: Symbol,
               type: Symbol,
               audio: OpenAI::Realtime::RealtimeSessionCreateResponse::Audio,
+              expires_at: Integer,
               include:
                 T::Array[
                   OpenAI::Realtime::RealtimeSessionCreateResponse::Include::TaggedSymbol
@@ -403,6 +425,7 @@ module OpenAI
                   OpenAI::Realtime::RealtimeSessionCreateResponse::OutputModality::TaggedSymbol
                 ],
               prompt: T.nilable(OpenAI::Responses::ResponsePrompt),
+              reasoning: OpenAI::Realtime::RealtimeReasoning,
               tool_choice:
                 OpenAI::Realtime::RealtimeSessionCreateResponse::ToolChoice::Variants,
               tools:
@@ -539,14 +562,6 @@ module OpenAI
             end
             attr_writer :noise_reduction
 
-            # Configuration for input audio transcription, defaults to off and can be set to
-            # `null` to turn off once on. Input audio transcription is not native to the
-            # model, since the model consumes audio directly. Transcription runs
-            # asynchronously through
-            # [the /audio/transcriptions endpoint](https://platform.openai.com/docs/api-reference/audio/createTranscription)
-            # and should be treated as guidance of input audio content rather than precisely
-            # what the model heard. The client can optionally set the language and prompt for
-            # transcription, these offer additional guidance to the transcription service.
             sig { returns(T.nilable(OpenAI::Realtime::AudioTranscription)) }
             attr_reader :transcription
 
@@ -570,6 +585,9 @@ module OpenAI
             # trails off with "uhhm", the model will score a low probability of turn end and
             # wait longer for the user to continue speaking. This can be useful for more
             # natural conversations, but may have a higher latency.
+            #
+            # For `gpt-realtime-whisper` transcription sessions, turn detection must be set to
+            # `null`; VAD is not supported.
             sig do
               returns(
                 T.nilable(
@@ -608,14 +626,6 @@ module OpenAI
               # detection accuracy (reducing false positives) and model performance by improving
               # perception of the input audio.
               noise_reduction: nil,
-              # Configuration for input audio transcription, defaults to off and can be set to
-              # `null` to turn off once on. Input audio transcription is not native to the
-              # model, since the model consumes audio directly. Transcription runs
-              # asynchronously through
-              # [the /audio/transcriptions endpoint](https://platform.openai.com/docs/api-reference/audio/createTranscription)
-              # and should be treated as guidance of input audio content rather than precisely
-              # what the model heard. The client can optionally set the language and prompt for
-              # transcription, these offer additional guidance to the transcription service.
               transcription: nil,
               # Configuration for turn detection, ether Server VAD or Semantic VAD. This can be
               # set to `null` to turn off, in which case the client must manually trigger model
@@ -630,6 +640,9 @@ module OpenAI
               # trails off with "uhhm", the model will score a low probability of turn end and
               # wait longer for the user to continue speaking. This can be useful for more
               # natural conversations, but may have a higher latency.
+              #
+              # For `gpt-realtime-whisper` transcription sessions, turn detection must be set to
+              # `null`; VAD is not supported.
               turn_detection: nil
             )
             end
@@ -717,6 +730,9 @@ module OpenAI
             # trails off with "uhhm", the model will score a low probability of turn end and
             # wait longer for the user to continue speaking. This can be useful for more
             # natural conversations, but may have a higher latency.
+            #
+            # For `gpt-realtime-whisper` transcription sessions, turn detection must be set to
+            # `null`; VAD is not supported.
             module TurnDetection
               extend OpenAI::Internal::Type::Union
 
@@ -1328,6 +1344,11 @@ module OpenAI
               :"gpt-realtime-1.5",
               OpenAI::Realtime::RealtimeSessionCreateResponse::Model::TaggedSymbol
             )
+          GPT_REALTIME_2 =
+            T.let(
+              :"gpt-realtime-2",
+              OpenAI::Realtime::RealtimeSessionCreateResponse::Model::TaggedSymbol
+            )
           GPT_REALTIME_2025_08_28 =
             T.let(
               :"gpt-realtime-2025-08-28",