omniai-openai 3.1.2 → 3.3.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 6c84d3becf5f93c58078e182e156861262bafc112e59ccb2c67d6ff08d0bb399
-  data.tar.gz: 31cb3de11d80088728a2f80193903f2740db4fd13c058b72db780561b25fc317
+  metadata.gz: 3d8e230669e337e7f48a0494bf3f7535f7d6d4a990ca2ed63c7929d76c0266a1
+  data.tar.gz: 5a5e9fe6083dab2b2f028d1ed56569bdc2a8cf4bc9b3d800da0809d6340aaf25
 SHA512:
-  metadata.gz: 8b789a962479764a1de021ee658cf27d82550e8d2699777bf7cc8294ffe099c9748ec4411d2296ece309e3dbd039685d0d2da0ed1f4380c573180fe042a7aa94
-  data.tar.gz: 180e1daaebcddb697f9e63d2b41e2adcd25a3e22e2845a184d43669497d415e350afeced8aa8cc99ec55c19dd67233685d034e138e80dcc0a26b7a089d216ca0
+  metadata.gz: 67d27b500204debfaa90b5da1eae2b4b6b398f4567c76ad361799930f958cedc3b720c2b60a03026651cfd7e036d8e5320623e6cdf0d91bfb12ca97a87f33668
+  data.tar.gz: 25990d1321f578a00ed518417d0f2d9c8725c4ba798b34f37bde7fea4ac96240eae146f0c3ba31a2d7fa5c37aef087ec3f5b5e3e5173e988014de0caa2fcb962

data/README.md

@@ -76,6 +76,18 @@ client  = OmniAI::OpenAI::Client.new(
             api_prefix: '/api')
 ```
 
+#### Usage with OpenAI-compatible gateways
+
+Private gateways and governed AI control planes can also be configured with a
+custom `host` while keeping application code on OmniAI's OpenAI provider:
+
+```ruby
+client = OmniAI::OpenAI::Client.new(
+  host: ENV.fetch('OPENAI_HOST', 'https://api.openai.com'),
+  api_key: ENV.fetch('OPENAI_API_KEY')
+)
+```
+
 ### Chat
 
 A chat completion is generated by passing in a simple text prompt:
@@ -97,13 +109,16 @@ completion.content # 'The capital of Canada is Ottawa.'
 
 #### Model
 
-`model` takes an optional string (default is `gpt-4o`):
+`model` takes an optional string (default is `gpt-5.2`):
 
 ```ruby
-completion = client.chat('How fast is a cheetah?', model: OmniAI::OpenAI::Chat::Model::GPT_3_5_TURBO)
+completion = client.chat('How fast is a cheetah?', model: OmniAI::OpenAI::Chat::Model::GPT_5_5)
 completion.content # 'A cheetah can reach speeds over 100 km/h.'
 ```
 
+Note that `temperature` is not supported by every model (e.g. `gpt-5`, `gpt-5.5` and the `o`-series). It is
+omitted from the request for those models rather than sent and rejected — see `TEMPERATURE_UNSUPPORTED_MODELS`.
+
 [OpenAI API Reference `model`](https://platform.openai.com/docs/api-reference/chat/create#chat-create-model)
 
 #### Temperature
@@ -148,20 +163,17 @@ JSON.parse(completion.content) # { "name": "Ringo" }
 
 #### Reasoning
 
-OpenAI o1 and o3 models support reasoning, which provides a summary of the model's thought process.
+OpenAI reasoning models (GPT-5 and later) can return a summary of the model's thought process.
 
 ```ruby
 # Enable reasoning with unified thinking API
-response = client.chat("What is 25 * 25?", model: "o3-mini", thinking: true)
+response = client.chat("What is 25 * 25?", model: "gpt-5.6-sol", thinking: true)
 
 # Or use OpenAI-specific reasoning options
-response = client.chat("What is 25 * 25?", model: "o3-mini", reasoning: { effort: "high", summary: "auto" })
+response = client.chat("What is 25 * 25?", model: "gpt-5.6-sol", reasoning: { effort: "high", summary: "auto" })
 ```
 
-**Reasoning Effort Levels:**
-- `low` - Minimal reasoning
-- `medium` - Balanced reasoning
-- `high` - Maximum reasoning effort
+**Reasoning Effort Levels** (`OmniAI::OpenAI::Chat::ReasoningEffort`): `none`, `minimal`, `low`, `medium`, `high`, `xhigh`, `max`. Support varies by model: GPT-6 models don't accept `minimal`, and `gpt-6-astra` and `gpt-6.1-sol` also reject `none`. Check the [model pages](https://developers.openai.com/api/docs/models/all).
 
 #### Accessing Reasoning Content
 
@@ -179,11 +191,28 @@ end
 #### Streaming with Reasoning
 
 ```ruby
-client.chat("What are the prime factors of 1234567?", model: "o3-mini", thinking: true, stream: $stdout)
+client.chat("What are the prime factors of 1234567?", model: "gpt-5.6-sol", thinking: true, stream: $stdout)
 ```
 
 [OpenAI API Reference `reasoning`](https://platform.openai.com/docs/guides/reasoning)
 
+#### Other Responses API Options
+
+Any option that is not modelled explicitly is forwarded to the Responses API verbatim, so parameters the gem
+does not wrap are still reachable:
+
+```ruby
+completion = client.chat('Summarize this.', model: 'gpt-5.5', max_output_tokens: 512, store: false)
+
+# continue a stored conversation
+completion = client.chat('And the next one?', previous_response_id: 'resp_123')
+```
+
+This covers `max_output_tokens`, `previous_response_id`, `store`, `parallel_tool_calls`, `metadata`,
+`truncation`, `top_p`, `service_tier`, and `prompt_cache_key`, among others.
+
+[OpenAI API Reference `responses`](https://platform.openai.com/docs/api-reference/responses/create)
+
 ### Transcribe
 
 A transcription is generated by passing in a path to a file:
@@ -269,10 +298,10 @@ client.speak('She sells seashells by the seashore.', voice: OmniAI::OpenAI::Spea
 
 #### Model
 
-`model` is optional and must be either `tts-1` or `tts-1-hd` (default):
+`model` is optional and is one of `gpt-4o-mini-tts` (default), `tts-1` or `tts-1-hd` (both shut down 2027-01-06):
 
 ```ruby
-client.speak('I saw a kitten eating chicken in the kitchen.', format: OmniAI::OpenAI::Speak::Model::TTS_1)
+client.speak('I saw a kitten eating chicken in the kitchen.', model: OmniAI::OpenAI::Speak::Model::GPT_4O_MINI_TTS)
 ```
 
 [OpenAI API Refernce `model`](https://platform.openai.com/docs/api-reference/audio/createSpeech#audio-createspeech-model)

data/lib/omniai/openai/chat/usage_serializer.rb

@@ -0,0 +1,34 @@
+# frozen_string_literal: true
+
+module OmniAI
+  module OpenAI
+    class Chat
+      # Overrides usage deserialize to read OpenAI's reasoning breakdown.
+      module UsageSerializer
+        # OpenAI already counts reasoning inside `output_tokens` and reports the breakdown alongside it, so the
+        # breakdown is read into `thinking_tokens` and the output count is left exactly as reported. Adding the two
+        # together would double count.
+        #
+        # This gem targets the Responses API (`/responses`), whose usage object reports the breakdown at
+        # `output_tokens_details.reasoning_tokens`. The Chat Completions vocabulary from the previous API
+        # generation (`completion_tokens_details`) is deliberately not read — this gem never receives it.
+        #
+        # @param data [Hash]
+        # @return [OmniAI::Chat::Usage]
+        def self.deserialize(data, *)
+          # Deserialize without a context so the generic flat parse runs rather than recursing into this method.
+          usage = OmniAI::Chat::Usage.deserialize(data)
+
+          # Only overwrite when the vendor container is actually present. A payload produced by `Usage#serialize`
+          # carries base's own `thinking_tokens` key and no `output_tokens_details`, so assigning unconditionally
+          # would clobber a correctly-parsed value with nil and break the round-trip. `unless nil?` rather than
+          # `||=`, so a reported zero from the wire still wins over base's nil.
+          reasoning_tokens = data.dig("output_tokens_details", "reasoning_tokens")
+          usage.thinking_tokens = reasoning_tokens unless reasoning_tokens.nil?
+
+          usage
+        end
+      end
+    end
+  end
+end

data/lib/omniai/openai/chat.rb

@@ -20,9 +20,12 @@ module OmniAI
 
       module ReasoningEffort
         NONE = "none"
+        MINIMAL = "minimal"
         LOW = "low"
         MEDIUM = "medium"
         HIGH = "high"
+        XHIGH = "xhigh"
+        MAX = "max"
       end
 
       module VerbosityText
@@ -32,28 +35,92 @@ module OmniAI
       end
 
       module Model
+        GPT_6_1_SOL = "gpt-6.1-sol"
+        GPT_6_ASTRA = "gpt-6-astra"
+        GPT_6_SOL = "gpt-6-sol"
+        GPT_6_LUNA = "gpt-6-luna"
+        GPT_5_6_LUNA = "gpt-5.6-luna"
+        GPT_5_6_SOL = "gpt-5.6-sol"
+        GPT_5_6_TERRA = "gpt-5.6-terra"
+        GPT_5_5 = "gpt-5.5"
+        GPT_5_5_PRO = "gpt-5.5-pro"
+        GPT_5_4 = "gpt-5.4"
+        GPT_5_4_MINI = "gpt-5.4-mini"
+        # @deprecated OpenAI shuts this down 2027-04-01; use GPT_6_LUNA.
+        GPT_5_4_NANO = "gpt-5.4-nano"
+        GPT_5_4_PRO = "gpt-5.4-pro"
+        # @deprecated OpenAI shuts this down 2027-04-01; use GPT_6_SOL.
+        GPT_5_3_CODEX = "gpt-5.3-codex"
         GPT_5_2 = "gpt-5.2"
+        GPT_5_2_PRO = "gpt-5.2-pro"
+        # @deprecated OpenAI shuts this down 2027-04-01; use GPT_6_SOL.
         GPT_5_1 = "gpt-5.1"
+        # @deprecated OpenAI shuts down its gpt-5-2025-08-07 snapshot 2026-12-11; use GPT_5_6_SOL.
         GPT_5 = "gpt-5"
+        # @deprecated OpenAI shuts down its gpt-5-mini-2025-08-07 snapshot 2026-12-11; use GPT_5_6_TERRA.
         GPT_5_MINI = "gpt-5-mini"
+        # @deprecated OpenAI shuts down its gpt-5-nano-2025-08-07 snapshot 2026-12-11; use GPT_5_6_LUNA.
         GPT_5_NANO = "gpt-5-nano"
+        # @deprecated OpenAI shuts down its gpt-5-pro-2025-10-06 snapshot 2026-12-11; use GPT_5_6_SOL.
+        GPT_5_PRO = "gpt-5-pro"
         GPT_4_1 = "gpt-4.1"
+        # @deprecated OpenAI shuts this down 2026-10-23; use GPT_5_6_LUNA.
         GPT_4_1_NANO = "gpt-4.1-nano"
         GPT_4_1_MINI = "gpt-4.1-mini"
         GPT_4O = "gpt-4o"
         GPT_4O_MINI = "gpt-4o-mini"
+        # @deprecated OpenAI shuts this down 2026-10-23; use GPT_5_6_SOL.
         GPT_4 = "gpt-4"
+        # @deprecated OpenAI shuts this down 2026-10-23; use GPT_5_6_SOL.
         GPT_4_TURBO = "gpt-4-turbo"
+        # @deprecated OpenAI shuts this down 2026-10-23; use GPT_5_6_TERRA.
         GPT_3_5_TURBO = "gpt-3.5-turbo"
+        # @deprecated OpenAI shut this down 2025-10-27; use GPT_5_6_SOL.
         O1_MINI = "o1-mini"
+        # @deprecated OpenAI shuts this down 2026-10-23; use GPT_5_6_SOL.
         O3_MINI = "o3-mini"
+        # @deprecated OpenAI shuts this down 2026-10-23; use GPT_5_6_TERRA.
         O4_MINI = "o4-mini"
+        # @deprecated OpenAI shuts this down 2026-10-23; use GPT_5_6_SOL.
         O1 = "o1"
+        # @deprecated OpenAI shuts down its o3-2025-04-16 snapshot 2026-12-11; use GPT_5_6_SOL.
         O3 = "o3"
       end
 
       DEFAULT_MODEL = Model::GPT_5_2
 
+      # Models that reject `temperature` on the Responses API.
+      #
+      # OpenAI's support here is not patterned by version, so this cannot be a prefix rule: `gpt-5` and its
+      # mini/nano/pro variants reject `temperature`, `gpt-5.1` through `gpt-5.4` accept it, and `gpt-5.5`
+      # onwards rejects it again. Every entry below was verified live against the API on 2026-08-16 by issuing
+      # a request with `temperature` and checking for `Unsupported parameter`.
+      #
+      # Models absent from this list pass `temperature` through unchanged. That is deliberate: an unlisted
+      # model that rejects it surfaces a loud API error the caller can act on, whereas over-applying the guard
+      # would silently discard a caller's temperature.
+      TEMPERATURE_UNSUPPORTED_MODELS = [
+        Model::GPT_5_6_LUNA,
+        Model::GPT_5_6_SOL,
+        Model::GPT_5_6_TERRA,
+        Model::GPT_5_5,
+        Model::GPT_5_5_PRO,
+        Model::GPT_5_4_PRO,
+        Model::GPT_5_2_PRO,
+        Model::GPT_5,
+        Model::GPT_5_MINI,
+        Model::GPT_5_NANO,
+        Model::GPT_5_PRO,
+        Model::O1_MINI,
+        Model::O1,
+        Model::O3,
+        Model::O3_MINI,
+        Model::O4_MINI,
+      ].freeze
+
+      # Keys in `@options` consumed by dedicated builders below rather than sent verbatim.
+      RESERVED_OPTIONS = %i[text reasoning thinking].freeze
+
       # @return [Context]
       CONTEXT = Context.build do |context|
         context.serializers[:choice] = ChoiceSerializer.method(:serialize)
@@ -65,6 +132,7 @@ module OmniAI
         context.serializers[:response] = ResponseSerializer.method(:serialize)
         context.deserializers[:response] = ResponseSerializer.method(:deserialize)
         context.deserializers[:content] = ContentSerializer.method(:deserialize)
+        context.deserializers[:usage] = UsageSerializer.method(:deserialize)
         context.serializers[:file] = FileSerializer.method(:serialize)
         context.serializers[:url] = URLSerializer.method(:serialize)
         context.serializers[:tool] = ToolSerializer.method(:serialize)
@@ -106,36 +174,38 @@ module OmniAI
         parts.join("\n\n")
       end
 
+      # @return [String]
+      def model
+        @model || DEFAULT_MODEL
+      end
+
       # @return [Float, nil]
       def temperature
         return if @temperature.nil?
-
-        return if [
-          Model::GPT_5,
-          Model::GPT_5_MINI,
-          Model::GPT_5_NANO,
-          Model::GPT_5_1,
-          Model::GPT_5_2,
-          Model::O1_MINI,
-          Model::O1,
-          Model::O3_MINI,
-        ].include?(@model)
+        return if TEMPERATURE_UNSUPPORTED_MODELS.include?(model)
 
         @temperature
       end
 
+      # Unrecognized options are passed through verbatim so callers can reach Responses API parameters the gem
+      # does not model explicitly (e.g. `max_output_tokens`, `previous_response_id`, `store`,
+      # `parallel_tool_calls`, `metadata`, `truncation`, `top_p`). Keys built by dedicated methods are excluded
+      # so they are not sent twice, and the explicit keys below win over any same-named passthrough.
+      #
       # @return [Hash]
       def payload
-        OmniAI::OpenAI.config.chat_options.merge({
-          instructions:,
-          input:,
-          model: @model || DEFAULT_MODEL,
-          stream: stream? || nil,
-          temperature:,
-          tools:,
-          text:,
-          reasoning:,
-        }).compact
+        OmniAI::OpenAI.config.chat_options
+          .merge(@options.except(*RESERVED_OPTIONS))
+          .merge({
+            instructions:,
+            input:,
+            model:,
+            stream: stream? || nil,
+            temperature:,
+            tools:,
+            text:,
+            reasoning:,
+          }).compact
       end
 
       # @return [Array<Hash>]

data/lib/omniai/openai/speak.rb

@@ -5,22 +5,28 @@ module OmniAI
     # An OpenAI transcribe implementation.
     class Speak < OmniAI::Speak
       module Model
+        # @deprecated OpenAI shuts this down 2027-01-06.
         TTS_1 = "tts-1"
+        # @deprecated OpenAI shuts this down 2027-01-06.
         TTS_1_HD = "tts-1-hd"
+        # @deprecated OpenAI shuts down both its snapshots 2027-01-06 (the current DEFAULT_MODEL).
         GPT_4O_MINI_TTS = "gpt-4o-mini-tts"
       end
 
       module Voice
         ALLOY = "alloy" # https://platform.openai.com/docs/guides/text-to-speech/alloy
         ASH = "ash" # https://platform.openai.com/docs/guides/text-to-speech/ash
-        BALLARD = "ballard" # https://platform.openai.com/docs/guides/text-to-speech/ballard
+        BALLAD = "ballad" # https://platform.openai.com/docs/guides/text-to-speech/ballad
+        CEDAR = "cedar" # https://platform.openai.com/docs/guides/text-to-speech/cedar
         CORAL = "coral" # https://platform.openai.com/docs/guides/text-to-speech/coral
         ECHO = "echo" # https://platform.openai.com/docs/guides/text-to-speech/echo
         FABLE = "fable" # https://platform.openai.com/docs/guides/text-to-speech/fable
+        MARIN = "marin" # https://platform.openai.com/docs/guides/text-to-speech/marin
         NOVA = "nova" # https://platform.openai.com/docs/guides/text-to-speech/nova
         ONYX = "onyx" # https://platform.openai.com/docs/guides/text-to-speech/onyx
         SAGE = "sage" # https://platform.openai.com/docs/guides/text-to-speech/sage
         SHIMMER = "shimmer" # https://platform.openai.com/docs/guides/text-to-speech/shimmer
+        VERSE = "verse" # https://platform.openai.com/docs/guides/text-to-speech/verse
       end
 
       DEFAULT_MODEL = Model::GPT_4O_MINI_TTS

data/lib/omniai/openai/transcribe.rb

@@ -5,9 +5,17 @@ module OmniAI
     # An OpenAI transcribe implementation.
     class Transcribe < OmniAI::Transcribe
       module Model
+        # @deprecated OpenAI shuts this down 2027-02-26; use GPT_TRANSCRIBE.
         WHISPER_1 = "whisper-1"
+        # @deprecated OpenAI shuts this down 2027-02-26; use GPT_TRANSCRIBE.
         GPT_4O_TRANSCRIBE = "gpt-4o-transcribe"
-        GPT_4O_MINI_TRANSCRIBE = "gpt-4-0-mini-transcribe"
+        # @deprecated OpenAI shuts this down 2027-02-26; use GPT_TRANSCRIBE.
+        GPT_4O_TRANSCRIBE_DIARIZE = "gpt-4o-transcribe-diarize"
+        # @deprecated OpenAI shuts this down 2027-02-26; use GPT_TRANSCRIBE.
+        GPT_4O_MINI_TRANSCRIBE = "gpt-4o-mini-transcribe"
+        GPT_TRANSCRIBE = "gpt-transcribe"
+        GPT_LIVE_TRANSCRIBE = "gpt-live-transcribe"
+        # @deprecated OpenAI shuts this down 2027-02-26; use GPT_TRANSCRIBE.
         WHISPER = WHISPER_1
       end
 

data/lib/omniai/openai/version.rb

@@ -2,6 +2,6 @@
 
 module OmniAI
   module OpenAI
-    VERSION = "3.1.2"
+    VERSION = "3.3.0"
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: omniai-openai
 version: !ruby/object:Gem::Version
-  version: 3.1.2
+  version: 3.3.0
 platform: ruby
 authors:
 - Kevin Sylvestre
@@ -29,14 +29,14 @@ dependencies:
     requirements:
     - - "~>"
       - !ruby/object:Gem::Version
-        version: '3.0'
+        version: '3.8'
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - "~>"
       - !ruby/object:Gem::Version
-        version: '3.0'
+        version: '3.8'
 - !ruby/object:Gem::Dependency
   name: openssl
   requirement: !ruby/object:Gem::Requirement
@@ -89,6 +89,7 @@ files:
 - lib/omniai/openai/chat/tool_call_serializer.rb
 - lib/omniai/openai/chat/tool_serializer.rb
 - lib/omniai/openai/chat/url_serializer.rb
+- lib/omniai/openai/chat/usage_serializer.rb
 - lib/omniai/openai/client.rb
 - lib/omniai/openai/config.rb
 - lib/omniai/openai/embed.rb