diff --git a/config/llm.yml b/config/llm.yml
index 29f621c7d..83cd9355f 100644
--- a/config/llm.yml
+++ b/config/llm.yml
@@ -126,6 +126,27 @@ features:
   pdf_faq_generation:
     models: [gpt-4.1-mini, gpt-5-mini, gpt-4.1, gpt-5.1, gpt-5.2]
     default: gpt-4.1-mini
+  help_center_article_generation:
+    models:
+      [
+        gpt-4.1-mini,
+        gpt-5-mini,
+        gpt-4.1,
+        gpt-5.1,
+        gpt-5.2,
+        claude-haiku-4.5,
+        claude-sonnet-4.5,
+        gemini-3-flash,
+        gemini-3-pro,
+      ]
+    default: gpt-5.2
+  onboarding_content_generation:
+    models:
+      [gpt-4.1, gpt-4.1-mini, gpt-5-mini, gpt-5.1, gpt-5.2]
+    default: gpt-4.1
+  help_center_query_translation:
+    models: [gpt-4.1-nano, gpt-4.1-mini, gpt-5-mini]
+    default: gpt-4.1-nano
   audio_transcription:
     models: [whisper-1]
     default: whisper-1
diff --git a/enterprise/app/services/captain/llm/article_translation_service.rb b/enterprise/app/services/captain/llm/article_translation_service.rb
index 5db26088e..fab0f934c 100644
--- a/enterprise/app/services/captain/llm/article_translation_service.rb
+++ b/enterprise/app/services/captain/llm/article_translation_service.rb
@@ -6,7 +6,7 @@ class Captain::Llm::ArticleTranslationService < Captain::BaseTaskService
   def perform
     raise ArgumentError, "Invalid type: #{type}" unless TYPES.include?(type)
 
-    response = make_api_call(model: translation_model, messages: messages)
+    response = make_api_call(feature: 'help_center_article_generation', messages: messages)
     return response if response[:error]
 
     response.merge(message: response[:message].strip)
@@ -33,10 +33,6 @@ class Captain::Llm::ArticleTranslationService < Captain::BaseTaskService
     @llm_credential ||= system_llm_credential
   end
 
-  def translation_model
-    @translation_model ||= InstallationConfig.find_by(name: 'CAPTAIN_OPEN_AI_MODEL')&.value.presence || GPT_MODEL
-  end
-
   def title_system_prompt
     <<~SYSTEM_PROMPT_MESSAGE
       You are a professional translator.
diff --git a/enterprise/app/services/captain/llm/article_writer_service.rb b/enterprise/app/services/captain/llm/article_writer_service.rb
index b94027248..73b49b0ed 100644
--- a/enterprise/app/services/captain/llm/article_writer_service.rb
+++ b/enterprise/app/services/captain/llm/article_writer_service.rb
@@ -6,7 +6,7 @@ class Captain::Llm::ArticleWriterService < Captain::BaseTaskService
   pattr_initialize [:account!, :source_pages!, { hint_title: nil }]
 
   def perform
-    response = make_api_call(model: writer_model, messages: messages, schema: RESPONSE_SCHEMA)
+    response = make_api_call(feature: 'help_center_article_generation', messages: messages, schema: RESPONSE_SCHEMA)
     return response if response[:error]
 
     response.merge(message: extract_payload(response[:message]))
@@ -92,10 +92,6 @@ class Captain::Llm::ArticleWriterService < Captain::BaseTaskService
     false
   end
 
-  def writer_model
-    'gpt-5.2'
-  end
-
   def build_follow_up_context?
     false
   end
diff --git a/enterprise/app/services/captain/llm/embedding_service.rb b/enterprise/app/services/captain/llm/embedding_service.rb
index 2fac54594..c78c70f23 100644
--- a/enterprise/app/services/captain/llm/embedding_service.rb
+++ b/enterprise/app/services/captain/llm/embedding_service.rb
@@ -6,7 +6,7 @@ class Captain::Llm::EmbeddingService
   def initialize(account_id: nil)
     Llm::Config.initialize!
     @account_id = account_id
-    @embedding_model = InstallationConfig.find_by(name: 'CAPTAIN_EMBEDDING_MODEL')&.value.presence || LlmConstants::DEFAULT_EMBEDDING_MODEL
+    @embedding_model = self.class.embedding_model
   end
 
   def self.embedding_model
diff --git a/enterprise/app/services/captain/llm/help_center_curation_service.rb b/enterprise/app/services/captain/llm/help_center_curation_service.rb
index 1f8b8acb2..37056dc25 100644
--- a/enterprise/app/services/captain/llm/help_center_curation_service.rb
+++ b/enterprise/app/services/captain/llm/help_center_curation_service.rb
@@ -9,7 +9,7 @@ class Captain::Llm::HelpCenterCurationService < Captain::BaseTaskService
   pattr_initialize [:account!, :links!]
 
   def perform
-    response = make_api_call(model: CURATION_MODEL, messages: messages, schema: RESPONSE_SCHEMA)
+    response = make_api_call(feature: 'onboarding_content_generation', model: CURATION_MODEL, messages: messages, schema: RESPONSE_SCHEMA)
     return response if response[:error]
 
     response.merge(message: extract_payload(response[:message]))
diff --git a/enterprise/app/services/captain/llm/translate_query_service.rb b/enterprise/app/services/captain/llm/translate_query_service.rb
index 3e05244d3..12fb841fa 100644
--- a/enterprise/app/services/captain/llm/translate_query_service.rb
+++ b/enterprise/app/services/captain/llm/translate_query_service.rb
@@ -1,6 +1,4 @@
 class Captain::Llm::TranslateQueryService < Captain::BaseTaskService
-  MODEL = 'gpt-4.1-nano'.freeze
-
   pattr_initialize [:account!]
 
   def translate(query, target_language:)
@@ -11,7 +9,7 @@ class Captain::Llm::TranslateQueryService < Captain::BaseTaskService
       { role: 'user', content: query }
     ]
 
-    response = make_api_call(model: MODEL, messages: messages)
+    response = make_api_call(feature: 'help_center_query_translation', messages: messages)
     return query if response[:error]
 
     response[:message].strip
diff --git a/enterprise/app/services/captain/llm/widget_tagline_service.rb b/enterprise/app/services/captain/llm/widget_tagline_service.rb
index 230c54165..155b10396 100644
--- a/enterprise/app/services/captain/llm/widget_tagline_service.rb
+++ b/enterprise/app/services/captain/llm/widget_tagline_service.rb
@@ -4,7 +4,7 @@ class Captain::Llm::WidgetTaglineService < Captain::BaseTaskService
   pattr_initialize [:account!]
 
   def perform
-    response = make_api_call(model: tagline_model, messages: messages, schema: RESPONSE_SCHEMA)
+    response = make_api_call(feature: 'onboarding_content_generation', messages: messages, schema: RESPONSE_SCHEMA)
     return response if response[:error]
 
     response.merge(message: extract_tagline(response[:message]))
@@ -68,10 +68,6 @@ class Captain::Llm::WidgetTaglineService < Captain::BaseTaskService
     false
   end
 
-  def tagline_model
-    @tagline_model ||= InstallationConfig.find_by(name: 'CAPTAIN_OPEN_AI_MODEL')&.value.presence || GPT_MODEL
-  end
-
   def build_follow_up_context?
     false
   end
diff --git a/enterprise/app/services/captain/onboarding/website_analyzer_service.rb b/enterprise/app/services/captain/onboarding/website_analyzer_service.rb
index 799b9de93..fb6bab33b 100644
--- a/enterprise/app/services/captain/onboarding/website_analyzer_service.rb
+++ b/enterprise/app/services/captain/onboarding/website_analyzer_service.rb
@@ -4,7 +4,7 @@ class Captain::Onboarding::WebsiteAnalyzerService < Llm::BaseAiService
   MAX_CONTENT_LENGTH = 8000
 
   def initialize(website_url)
-    super()
+    super(feature: 'onboarding_content_generation')
     @website_url = normalize_url(website_url)
     @website_content = nil
     @favicon_url = nil
diff --git a/enterprise/app/services/messages/audio_transcription_service.rb b/enterprise/app/services/messages/audio_transcription_service.rb
index 748bf1efa..93ed867f2 100644
--- a/enterprise/app/services/messages/audio_transcription_service.rb
+++ b/enterprise/app/services/messages/audio_transcription_service.rb
@@ -1,7 +1,7 @@
 class Messages::AudioTranscriptionService< Llm::LegacyBaseOpenAiService
   include Integrations::LlmInstrumentation
 
-  TRANSCRIPTION_MODEL = 'gpt-4o-mini-transcribe'.freeze
+  TRANSCRIPTION_MODEL = 'whisper-1'.freeze
   # OpenAI's transcription endpoint hard limit is 25 MB *decimal* (25_000_000), not
   # binary (25.megabytes = 26_214_400) — using the binary form leaks the 25.0–26.2 MB
   # range to the API as 413s. Long audio (~70+ min Opus) keeps the attachment but skips
@@ -15,6 +15,7 @@ class Messages::AudioTranscriptionService< Llm::LegacyBaseOpenAiService
     @attachment = attachment
     @message = attachment.message
     @account = message.account
+    @transcription_model = Llm::FeatureRouter.resolve(feature: 'audio_transcription', account: account)[:model]
   end
 
   def perform
@@ -81,7 +82,7 @@ class Messages::AudioTranscriptionService< Llm::LegacyBaseOpenAiService
       # behaviour across OpenAI transcription models.
       response = @client.audio.transcribe(
         parameters: {
-          model: TRANSCRIPTION_MODEL,
+          model: transcription_model,
           file: file,
           temperature: 0.0
         }
@@ -98,7 +99,7 @@ class Messages::AudioTranscriptionService< Llm::LegacyBaseOpenAiService
   def instrumentation_params(file_path)
     {
       span_name: 'llm.messages.audio_transcription',
-      model: TRANSCRIPTION_MODEL,
+      model: transcription_model,
       account_id: account&.id,
       feature_name: 'audio_transcription',
       file_path: file_path
@@ -127,4 +128,8 @@ class Messages::AudioTranscriptionService< Llm::LegacyBaseOpenAiService
       'x-mp3' => 'mp3'
     }.fetch(subtype, subtype)
   end
+
+  def transcription_model
+    @transcription_model || TRANSCRIPTION_MODEL
+  end
 end
diff --git a/lib/captain/base_task_service.rb b/lib/captain/base_task_service.rb
index d098b24af..cfeb4e427 100644
--- a/lib/captain/base_task_service.rb
+++ b/lib/captain/base_task_service.rb
@@ -59,7 +59,10 @@ class Captain::BaseTaskService
   def resolved_model(model:, feature:)
     return model if feature.blank?
 
-    Llm::FeatureRouter.resolve(feature: feature, account: account)[:model]
+    route = Llm::FeatureRouter.resolve(feature: feature, account: account)
+    return model if model.present? && route[:source] == :default
+
+    route[:model]
   end
 
   def execute_ruby_llm_request(model:, messages:, schema: nil, tools: [])
diff --git a/spec/enterprise/services/captain/llm/article_translation_service_spec.rb b/spec/enterprise/services/captain/llm/article_translation_service_spec.rb
index 1c0d83b65..0661a906a 100644
--- a/spec/enterprise/services/captain/llm/article_translation_service_spec.rb
+++ b/spec/enterprise/services/captain/llm/article_translation_service_spec.rb
@@ -17,6 +17,7 @@ RSpec.describe Captain::Llm::ArticleTranslationService do
 
     it 'returns the stripped translated title' do
       expect(service).to receive(:make_api_call) do |args|
+        expect(args[:feature]).to eq('help_center_article_generation')
         expect(args[:messages][0][:content]).to include('professional translator')
         expect(args[:messages][0][:content]).to include(target_language)
         expect(args[:messages][1][:content]).to eq('Getting Started')
diff --git a/spec/enterprise/services/captain/llm/embedding_service_spec.rb b/spec/enterprise/services/captain/llm/embedding_service_spec.rb
new file mode 100644
index 000000000..206ca147d
--- /dev/null
+++ b/spec/enterprise/services/captain/llm/embedding_service_spec.rb
@@ -0,0 +1,38 @@
+require 'rails_helper'
+
+RSpec.describe Captain::Llm::EmbeddingService, type: :service do
+  def configure_embedding_model(value)
+    InstallationConfig.find_or_initialize_by(name: 'CAPTAIN_EMBEDDING_MODEL').tap do |config|
+      config.value = value
+      config.locked = false
+      config.save!
+    end
+  end
+
+  describe '.embedding_model' do
+    it 'uses the installation embedding model when configured' do
+      configure_embedding_model('custom-embedding-model')
+
+      expect(described_class.embedding_model).to eq('custom-embedding-model')
+    end
+
+    it 'falls back to the default embedding model when the installation value is blank' do
+      configure_embedding_model('')
+
+      expect(described_class.embedding_model).to eq(LlmConstants::DEFAULT_EMBEDDING_MODEL)
+    end
+  end
+
+  describe '#get_embedding' do
+    let(:account) { create(:account) }
+    let(:embedding_response) { double('embedding_response', vectors: [0.1, 0.2]) } # rubocop:disable RSpec/VerifiedDoubles
+
+    it 'sends the installation embedding model to RubyLLM' do
+      configure_embedding_model('custom-embedding-model')
+
+      expect(RubyLLM).to receive(:embed).with('search text', model: 'custom-embedding-model').and_return(embedding_response)
+
+      expect(described_class.new(account_id: account.id).get_embedding('search text')).to eq([0.1, 0.2])
+    end
+  end
+end
diff --git a/spec/enterprise/services/messages/audio_transcription_service_spec.rb b/spec/enterprise/services/messages/audio_transcription_service_spec.rb
index 32752c2b2..ecc574501 100644
--- a/spec/enterprise/services/messages/audio_transcription_service_spec.rb
+++ b/spec/enterprise/services/messages/audio_transcription_service_spec.rb
@@ -3,7 +3,7 @@ require 'rails_helper'
 RSpec.describe Messages::AudioTranscriptionService, type: :service do
   let(:account) { create(:account, audio_transcriptions: true) }
   let(:conversation) { create(:conversation, account: account) }
-  let(:message) { create(:message, conversation: conversation) }
+  let(:message) { create(:message, account: account, conversation: conversation) }
   let(:attachment) { message.attachments.create!(account: account, file_type: :audio) }
 
   before do
@@ -101,4 +101,29 @@ RSpec.describe Messages::AudioTranscriptionService, type: :service do
       FileUtils.rm_f(temp_file_path) if temp_file_path.present?
     end
   end
+
+  describe '#transcribe_audio' do
+    let(:service) { described_class.new(attachment) }
+    let(:audio_api) { double('audio_api') } # rubocop:disable RSpec/VerifiedDoubles
+    let(:audio_file_path) { Rails.root.join('tmp/audio_transcription_service_spec.mp3').to_s }
+
+    before do
+      File.binwrite(audio_file_path, 'audio')
+      allow(service).to receive(:fetch_audio_file).and_return(audio_file_path)
+      allow(service).to receive(:update_transcription)
+      allow(service.client).to receive(:audio).and_return(audio_api)
+    end
+
+    after do
+      FileUtils.rm_f(audio_file_path)
+    end
+
+    it 'uses the audio transcription feature model' do
+      expect(audio_api).to receive(:transcribe).with(
+        parameters: hash_including(model: 'whisper-1', temperature: 0.0)
+      ).and_return({ 'text' => 'Audio transcript' })
+
+      expect(service.send(:transcribe_audio)).to eq('Audio transcript')
+    end
+  end
 end
diff --git a/spec/lib/captain/base_task_service_spec.rb b/spec/lib/captain/base_task_service_spec.rb
index 7b67ab94a..b24a5c49c 100644
--- a/spec/lib/captain/base_task_service_spec.rb
+++ b/spec/lib/captain/base_task_service_spec.rb
@@ -21,6 +21,7 @@ RSpec.describe Captain::BaseTaskService do
   let(:service) { test_service_class.new(account: account, conversation_display_id: conversation.display_id) }
 
   before do
+    InstallationConfig.where(name: 'CAPTAIN_OPEN_AI_API_KEY').destroy_all
     create(:installation_config, name: 'CAPTAIN_OPEN_AI_API_KEY', value: 'test-key')
     # Stub captain enabled check to allow OSS specs to test base functionality
     # without enterprise module interference
@@ -178,6 +179,26 @@ RSpec.describe Captain::BaseTaskService do
       service.send(:make_api_call, feature: 'editor', messages: messages)
     end
 
+    it 'uses the supplied model as a feature fallback when there is no account override' do
+      expect(mock_context).to receive(:chat).with(model: 'gpt-5.2').and_return(mock_chat)
+
+      service.send(:make_api_call, feature: 'document_faq_generation', model: 'gpt-5.2', messages: messages)
+    end
+
+    it 'uses the help center article generation feature default' do
+      expect(mock_context).to receive(:chat).with(model: 'gpt-5.2').and_return(mock_chat)
+
+      service.send(:make_api_call, feature: 'help_center_article_generation', messages: messages)
+    end
+
+    it 'prefers account overrides over supplied feature fallback models' do
+      account.update!(captain_models: { 'help_center_article_generation' => 'gpt-4.1' })
+
+      expect(mock_context).to receive(:chat).with(model: 'gpt-4.1').and_return(mock_chat)
+
+      service.send(:make_api_call, feature: 'help_center_article_generation', model: 'gpt-5.2', messages: messages)
+    end
+
     it 'returns formatted response with tokens' do
       result = service.send(:make_api_call, model: model, messages: messages)