diff --git a/CLAUDE.md b/CLAUDE.md
index 7c83d66d..40d1f8ef 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -15,7 +15,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 - **Start Backend**: `python run.py` (from backend/ directory)
 - **Backend Server**: Runs on port 5137 with Hypercorn ASGI server
 - **Database**: MongoDB with PyMongo
-- **Authentication**: JWT tokens via Flask-JWT-Extended
+- **Authentication**: Custom Quart-compatible JWT (not Flask-JWT-Extended)
 
 ## Testing
 **Python Backend**: After modifying any Python files:
@@ -26,9 +26,30 @@ python -c "from app import create_app; app = create_app()"  # Test app creation
 ```
 **Frontend**: Run `npm run build` to verify TypeScript compilation
 
+## Architecture Overview
+
+### Real-Time Communication
+The application uses Socket.IO for real-time WebSocket communication between frontend and backend:
+- **Backend**: `python-socketio` with `AsyncServer` wrapped in ASGI app
+- **Frontend**: `socket.io-client` managed via `WebSocketContext`
+- WebSocket manager (`websocket_manager_async.py`) handles room-based messaging for focus group sessions
+
+### Autonomous Conversation System
+Focus group sessions can run autonomously with AI-driven conversations:
+- `ai_runner_service.py` - Manages background task execution for autonomous mode
+- `autonomous_conversation_controller.py` - Orchestrates multi-persona conversations
+- `conversation_decision_service.py` - Determines next speaker and conversation flow
+- `conversation_context_service.py` - Maintains conversation state and history
+
+### LLM Integration
+Multi-model support through `llm_service.py`:
+- **Google Gemini** (`gemini-3-pro-preview`) - Default model
+- **OpenAI** (`gpt-4.1`, `gpt-5`) - Alternative models
+- Prompts are stored as markdown templates in `/backend/prompts/`
+
 ## Code Style Guidelines
 - **Imports**: Group imports by source (React, third-party, local)
-- **Types**: Use TypeScript. Project allows nullable types (`strictNullChecks: false`) 
+- **Types**: Use TypeScript. Project allows nullable types (`strictNullChecks: false`)
 - **Components**: Use functional components with hooks
 - **Naming**: Use PascalCase for components, camelCase for variables/functions
 - **Formatting**: Follow ESLint recommendations, focus on readability
@@ -41,12 +62,13 @@ python -c "from app import create_app; app = create_app()"  # Test app creation
 
 ## Project Stack
 **Frontend**: Vite, React 18, TypeScript, Tailwind CSS, shadcn-ui
-**Backend**: Flask 2.2.3, Hypercorn, PyMongo, JWT Extended
-**Key Libraries**: 
+**Backend**: Quart (async Flask), Hypercorn ASGI, PyMongo, python-socketio
+**Key Libraries**:
 - UI: Radix UI components, Lucide React icons
 - State: TanStack Query, React Hook Form with Zod validation
 - Routing: React Router DOM
-- AI/LLM: OpenAI, Google Generative AI
+- AI/LLM: OpenAI, Google Generative AI (genai)
+- Real-time: Socket.IO (client and server)
 - Charts: Recharts
 - Drag & Drop: DND Kit
 
@@ -58,15 +80,15 @@ python -c "from app import create_app; app = create_app()"  # Test app creation
 
 ## File Organization
 - **Backend Services**: `/backend/app/services/` - Business logic and AI integrations
-- **Backend Models**: `/backend/app/models/` - Data models (User, FocusGroup, Persona)
-- **Backend Routes**: `/backend/app/routes/` - API endpoints
-- **AI Prompts**: `/backend/prompts/` - LLM prompt templates
-- **Frontend Components**: 
+- **Backend Models**: `/backend/app/models/` - Data models (User, FocusGroup, Persona, Folder)
+- **Backend Routes**: `/backend/app/routes/` - API endpoints (auth, personas, focus-groups, ai-personas, folders, tasks)
+- **AI Prompts**: `/backend/prompts/` - LLM prompt templates (markdown files loaded by `prompt_loader.py`)
+- **Frontend Components**:
   - `/src/components/ui/` - Reusable shadcn-ui components
-  - `/src/components/focus-group-session/` - Focus group specific components
+  - `/src/components/focus-group-session/` - Focus group session UI (DiscussionPanel, ParticipantPanel, ThemesPanel, etc.)
   - `/src/components/persona/` - Persona management components
 - **Types**: `/src/types/` - TypeScript type definitions
-- **Contexts**: `/src/contexts/` - React context providers
+- **Contexts**: `/src/contexts/` - React context providers (AuthContext, WebSocketContext, NavigationContext)
 
 ## Environment Configuration
 
diff --git a/backend/app/models/focus_group.py b/backend/app/models/focus_group.py
index bc819754..4ffecdbc 100644
--- a/backend/app/models/focus_group.py
+++ b/backend/app/models/focus_group.py
@@ -82,7 +82,7 @@ class FocusGroup:
         
         # Set default LLM model if not provided
         if "llm_model" not in focus_group_data:
-            focus_group_data["llm_model"] = "gemini-2.5-pro"
+            focus_group_data["llm_model"] = "gemini-3-pro-preview"
         
         # Set default GPT-5 parameters if not provided
         if "reasoning_effort" not in focus_group_data:
diff --git a/backend/app/routes/ai_personas.py b/backend/app/routes/ai_personas.py
index 79d5aba1..3f1af550 100644
--- a/backend/app/routes/ai_personas.py
+++ b/backend/app/routes/ai_personas.py
@@ -67,7 +67,7 @@ async def generate_basic_profiles():
         temperature = 1.0
     
     customer_data_session_id = data.get('customer_data_session_id')  # Optional parameter
-    llm_model = data.get('llm_model', 'gemini-2.5-pro')  # Optional parameter with default
+    llm_model = data.get('llm_model', 'gemini-3-pro-preview')  # Optional parameter with default
     
     try:
         # Register current task for cancellation
@@ -197,7 +197,7 @@ async def complete_and_save_persona():
         temperature = 1.0
     
     customer_data_session_id = data.get('customer_data_session_id')  # Optional parameter
-    llm_model = data.get('llm_model', 'gemini-2.5-pro')  # Optional parameter with default
+    llm_model = data.get('llm_model', 'gemini-3-pro-preview')  # Optional parameter with default
     
     # Get persona name for logging
     persona_name = basic_profile.get('name', 'Unknown')
@@ -816,7 +816,7 @@ async def batch_generate_summaries():
     if not (0 <= temperature <= 1.5):
         temperature = 1.0
     
-    llm_model = data.get('llm_model', 'gemini-2.5-pro')  # Optional parameter with default
+    llm_model = data.get('llm_model', 'gemini-3-pro-preview')  # Optional parameter with default
     
     # Log the request with model information
     print(f"🔄 Backend: Received batch-generate-summaries request for {len(persona_ids)} personas with model: {llm_model}")
@@ -1013,7 +1013,7 @@ async def generate_personas_full():
         "count": 5,
         "temperature": 0.8,
         "customer_data_session_id": "optional_session_id",
-        "llm_model": "gemini-2.5-pro",
+        "llm_model": "gemini-3-pro-preview",
         "target_folder_id": "optional_folder_id"
     }
     
@@ -1038,7 +1038,7 @@ async def generate_personas_full():
         temperature = 1.0
     
     customer_data_session_id = data.get('customer_data_session_id')
-    llm_model = data.get('llm_model', 'gemini-2.5-pro')
+    llm_model = data.get('llm_model', 'gemini-3-pro-preview')
     target_folder_id = data.get('target_folder_id')
     
     try:
diff --git a/backend/app/routes/focus_group_ai.py b/backend/app/routes/focus_group_ai.py
index 3c2df9df..7788f01a 100644
--- a/backend/app/routes/focus_group_ai.py
+++ b/backend/app/routes/focus_group_ai.py
@@ -84,7 +84,7 @@ async def generate_ai_response():
             import datetime
             log_msg = f"🤖 [{datetime.datetime.now()}] AI RESPONSE - Focus group keys: {list(focus_group.keys())}\n"
             log_msg += f"🤖 [{datetime.datetime.now()}] AI RESPONSE - Raw llm_model from DB: '{focus_group.get('llm_model')}' (type: {type(focus_group.get('llm_model'))})\n" 
-            log_msg += f"🤖 [{datetime.datetime.now()}] AI RESPONSE - Using model: {llm_model or 'default (gemini-2.5-pro)'} for focus group {focus_group_id}\n"
+            log_msg += f"🤖 [{datetime.datetime.now()}] AI RESPONSE - Using model: {llm_model or 'default (gemini-3-pro-preview)'} for focus group {focus_group_id}\n"
             with open('/tmp/focus_group_debug.log', 'a') as f:
                 f.write(log_msg)
                 f.flush()
@@ -93,7 +93,7 @@ async def generate_ai_response():
         
         current_app.logger.info(f"🔍 DEBUG: Focus group data keys: {list(focus_group.keys())}")
         current_app.logger.info(f"🔍 DEBUG: Raw llm_model value from DB: '{focus_group.get('llm_model')}' (type: {type(focus_group.get('llm_model'))})")
-        current_app.logger.info(f"🤖 Generating AI response using model: {llm_model or 'default (gemini-2.5-pro)'} for focus group {focus_group_id}")
+        current_app.logger.info(f"🤖 Generating AI response using model: {llm_model or 'default (gemini-3-pro-preview)'} for focus group {focus_group_id}")
         
         # Validate persona exists
         persona = await Persona.find_by_id(persona_id)
diff --git a/backend/app/routes/personas.py b/backend/app/routes/personas.py
index c30fbd7e..c5acd216 100644
--- a/backend/app/routes/personas.py
+++ b/backend/app/routes/personas.py
@@ -173,7 +173,7 @@ async def modify_persona_with_ai(persona_id):
     
     Request body should include:
     - modification_prompt: Natural language description of desired changes
-    - llm_model: Model to use (defaults to 'gemini-2.5-pro')
+    - llm_model: Model to use (defaults to 'gemini-3-pro-preview')
     - reasoning_effort: For GPT-5 (minimal, low, medium, high)
     - verbosity: For GPT-5 (low, medium, high)
     - preview_only: If true, returns modified data without saving to database (defaults to false)
@@ -188,7 +188,7 @@ async def modify_persona_with_ai(persona_id):
         if not modification_prompt:
             return jsonify({"error": "modification_prompt is required"}), 400
         
-        llm_model = request_data.get('llm_model', 'gemini-2.5-pro')
+        llm_model = request_data.get('llm_model', 'gemini-3-pro-preview')
         reasoning_effort = request_data.get('reasoning_effort', 'medium')
         verbosity = request_data.get('verbosity', 'medium')
         preview_only = request_data.get('preview_only', False)
diff --git a/backend/app/services/ai_persona_service.py b/backend/app/services/ai_persona_service.py
index eee84842..62e0dc43 100644
--- a/backend/app/services/ai_persona_service.py
+++ b/backend/app/services/ai_persona_service.py
@@ -218,7 +218,7 @@ async def _generate_basic_personas_attempt(
 
             # Log the LLM API call with attempt number
             attempt_text = f" (attempt {attempt})" if attempt > 1 else ""
-            print(f"🤖 Backend: Making LLM API call to {llm_model or 'gemini-2.5-pro'} for basic persona generation{attempt_text}")
+            print(f"🤖 Backend: Making LLM API call to {llm_model or 'gemini-3-pro-preview'} for basic persona generation{attempt_text}")
 
             raw_response = await LLMService.generate_content(
                 prompt=final_prompt,
@@ -504,7 +504,7 @@ async def generate_persona(
             
             # Log the LLM API call
             persona_name = basic_persona.get('name', 'Unknown') if basic_persona else 'New Persona'
-            print(f"🤖 Backend: Making LLM API call to {llm_model or 'gemini-2.5-pro'} for detailed persona generation of '{persona_name}'")
+            print(f"🤖 Backend: Making LLM API call to {llm_model or 'gemini-3-pro-preview'} for detailed persona generation of '{persona_name}'")
             
             persona_data = await LLMService.generate_structured_response(
                 prompt=final_prompt,
@@ -589,7 +589,7 @@ async def generate_persona_summary(
             
             # Log the LLM API call
             persona_name = persona_data.get('name', 'Unknown')
-            print(f"🤖 Backend: Making LLM API call to {llm_model or 'gemini-2.5-pro'} for summary generation of '{persona_name}'")
+            print(f"🤖 Backend: Making LLM API call to {llm_model or 'gemini-3-pro-preview'} for summary generation of '{persona_name}'")
             
             raw_response = await LLMService.generate_content(
                 prompt=final_prompt,
@@ -694,7 +694,7 @@ async def generate_persona_download_summary(
             
             # Log the LLM API call
             persona_name = persona_data.get('name', 'Unknown')
-            print(f"🤖 Backend: Making LLM API call to {llm_model or 'gemini-2.5-pro'} for download summary of '{persona_name}'")
+            print(f"🤖 Backend: Making LLM API call to {llm_model or 'gemini-3-pro-preview'} for download summary of '{persona_name}'")
             
             # Generate the markdown content directly
             markdown_response = await LLMService.generate_content(
diff --git a/backend/app/services/autonomous_conversation_controller.py b/backend/app/services/autonomous_conversation_controller.py
index 5dbbd967..c0547f88 100644
--- a/backend/app/services/autonomous_conversation_controller.py
+++ b/backend/app/services/autonomous_conversation_controller.py
@@ -664,7 +664,7 @@ class AutonomousConversationController:
             llm_model = focus_group.get('llm_model')
             reasoning_effort = focus_group.get('reasoning_effort', 'medium')
             verbosity = focus_group.get('verbosity', 'medium')
-            self.logger.info(f"🤖 Autonomous conversation using model: {llm_model or 'default (gemini-2.5-pro)'} for focus group {self.focus_group_id}")
+            self.logger.info(f"🤖 Autonomous conversation using model: {llm_model or 'default (gemini-3-pro-preview)'} for focus group {self.focus_group_id}")
             
             # Get recent messages
             messages = await FocusGroup.get_messages(self.focus_group_id)
diff --git a/backend/app/services/focus_group_response_service.py b/backend/app/services/focus_group_response_service.py
index 858f52c7..67da7b3d 100644
--- a/backend/app/services/focus_group_response_service.py
+++ b/backend/app/services/focus_group_response_service.py
@@ -52,7 +52,7 @@ async def generate_persona_response(
         if llm_model == 'gpt-5':
             print(f"  - llm_model: {llm_model} (reasoning_effort: {reasoning_effort or 'medium'}, verbosity: {verbosity or 'medium'}) [using Responses API]")
         else:
-            print(f"  - llm_model: {llm_model or 'default (gemini-2.5-pro)'}")
+            print(f"  - llm_model: {llm_model or 'default (gemini-3-pro-preview)'}")
         
         # Import LLMService at the top to avoid scoping issues
         from app.services.llm_service import LLMService
diff --git a/backend/app/services/focus_group_service.py b/backend/app/services/focus_group_service.py
index d3ba0f2e..7060d76f 100644
--- a/backend/app/services/focus_group_service.py
+++ b/backend/app/services/focus_group_service.py
@@ -224,7 +224,7 @@ class FocusGroupService:
                                                 'content': question.get('content', 'No content')[:100] + '...'
                                             })
                             
-                            logger.info(f"=== CREATIVE REVIEW VALIDATION RESULTS (Model: {llm_model or 'gemini-2.5-pro'}) ===")
+                            logger.info(f"=== CREATIVE REVIEW VALIDATION RESULTS (Model: {llm_model or 'gemini-3-pro-preview'}) ===")
                             logger.info(f"Found {creative_review_count} creative_review activities for {len(uploaded_assets)} uploaded assets")
                             
                             if creative_review_activities:
@@ -236,7 +236,7 @@ class FocusGroupService:
                             # If no creative review activities were generated, retry with enhanced prompt
                             if creative_review_count == 0:
                                 logger.warning(f"❌ WARNING: No creative_review activities generated despite {len(uploaded_assets)} uploaded assets!")
-                                logger.warning(f"❌ This suggests {llm_model or 'gemini-2.5-pro'} is not following the creative asset instructions")
+                                logger.warning(f"❌ This suggests {llm_model or 'gemini-3-pro-preview'} is not following the creative asset instructions")
                                 
                                 # For GPT models, if this was already the enhanced prompt, we have a serious issue
                                 if llm_model and llm_model.startswith('gpt') and attempt < max_retries:
diff --git a/backend/app/services/key_theme_service.py b/backend/app/services/key_theme_service.py
index 2ffc3683..9eaa8370 100644
--- a/backend/app/services/key_theme_service.py
+++ b/backend/app/services/key_theme_service.py
@@ -41,7 +41,7 @@ class KeyThemeService:
         """
         logger = logging.getLogger(__name__)
         logger.info(f"Starting key theme generation for focus group {focus_group_id} with temperature {temperature}")
-        logger.info(f"Using LLM model: {llm_model or 'default (gemini-2.5-pro)'}")
+        logger.info(f"Using LLM model: {llm_model or 'default (gemini-3-pro-preview)'}")
         
         try:
             # Get the focus group
@@ -105,7 +105,7 @@ class KeyThemeService:
         """
         logger = logging.getLogger(__name__)
         logger.info(f"Beginning theme extraction from {len(messages)} messages")
-        logger.info(f"Theme extraction using LLM model: {llm_model or 'default (gemini-2.5-pro)'}")
+        logger.info(f"Theme extraction using LLM model: {llm_model or 'default (gemini-3-pro-preview)'}")
         
         try:
             # Load and prepare the prompt for the LLM
@@ -135,7 +135,7 @@ class KeyThemeService:
             
             for attempt in range(max_retries):
                 attempt_num = attempt + 1
-                logger.info(f"Attempt {attempt_num}/{max_retries}: Calling LLM ({llm_model or 'gemini-2.5-pro'}) for theme generation")
+                logger.info(f"Attempt {attempt_num}/{max_retries}: Calling LLM ({llm_model or 'gemini-3-pro-preview'}) for theme generation")
                 
                 try:
                     themes = await LLMService.generate_structured_array(
@@ -145,7 +145,7 @@ class KeyThemeService:
                         model_name=llm_model
                     )
                     
-                    logger.info(f"Attempt {attempt_num}/{max_retries}: LLM ({llm_model or 'gemini-2.5-pro'}) call successful, received {len(themes)} themes")
+                    logger.info(f"Attempt {attempt_num}/{max_retries}: LLM ({llm_model or 'gemini-3-pro-preview'}) call successful, received {len(themes)} themes")
                     
                     # Validate the response structure
                     validated_themes = []
@@ -175,7 +175,7 @@ class KeyThemeService:
                             
                             validated_themes.append(validated_theme)
                     
-                    logger.info(f"Theme generation completed successfully with {len(validated_themes)} validated themes using {llm_model or 'gemini-2.5-pro'}")
+                    logger.info(f"Theme generation completed successfully with {len(validated_themes)} validated themes using {llm_model or 'gemini-3-pro-preview'}")
                     return validated_themes
                     
                 except LLMServiceError as e:
diff --git a/backend/app/services/llm_service.py b/backend/app/services/llm_service.py
index 8acba2cb..3f676899 100644
--- a/backend/app/services/llm_service.py
+++ b/backend/app/services/llm_service.py
@@ -25,12 +25,12 @@ gemini_client = genai.Client(api_key=GEMINI_API_KEY)
 OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY', 'REDACTED_OPENAI_KEY')
 openai_client = AsyncOpenAI(api_key=OPENAI_API_KEY, timeout=600.0)
 
-# The default model we're using 
-DEFAULT_MODEL = "gemini-2.5-pro"
+# The default model we're using
+DEFAULT_MODEL = "gemini-3-pro-preview"
 
 # Supported models
 SUPPORTED_MODELS = {
-    'gemini-2.5-pro': 'gemini',
+    'gemini-3-pro-preview': 'gemini',
     'gpt-4.1': 'openai',
     'gpt-5': 'openai'
 }
diff --git a/backend/app/services/persona_modification_service.py b/backend/app/services/persona_modification_service.py
index bad46854..487681c7 100644
--- a/backend/app/services/persona_modification_service.py
+++ b/backend/app/services/persona_modification_service.py
@@ -134,7 +134,7 @@ class PersonaModificationService:
     async def modify_persona(
         persona_id: str,
         modification_prompt: str,
-        llm_model: str = 'gemini-2.5-pro',
+        llm_model: str = 'gemini-3-pro-preview',
         reasoning_effort: str = 'medium',
         verbosity: str = 'medium',
         max_retries: int = 3,
diff --git a/src/components/AIRecruiter.tsx b/src/components/AIRecruiter.tsx
index 0a8028a7..edf636fa 100644
--- a/src/components/AIRecruiter.tsx
+++ b/src/components/AIRecruiter.tsx
@@ -159,7 +159,7 @@ export default function AIRecruiter({ targetFolderId, targetFolderName }: AIRecr
         if (response.partial_success || (response.errors && response.errors.length > 0)) {
           // Some personas succeeded but others failed
           toast.success("Some personas generated successfully", {
-            description: `${personas.length} synthetic personas were created using ${values.llm_model || 'Gemini 2.5 Pro'}. ${response.errors?.length || 0} failed due to timeout or other errors.`,
+            description: `${personas.length} synthetic personas were created using ${values.llm_model || 'Gemini 3 Pro'}. ${response.errors?.length || 0} failed due to timeout or other errors.`,
             duration: 8000
           });
           
@@ -175,7 +175,7 @@ export default function AIRecruiter({ targetFolderId, targetFolderName }: AIRecr
         } else {
           // All personas succeeded
           toast.success("Personas generated and saved successfully", {
-            description: `${personas.length} synthetic personas have been created using ${values.llm_model || 'Gemini 2.5 Pro'} and saved ${targetFolderId ? `to the "${targetFolderName}" folder` : 'to the database'}.`
+            description: `${personas.length} synthetic personas have been created using ${values.llm_model || 'Gemini 3 Pro'} and saved ${targetFolderId ? `to the "${targetFolderName}" folder` : 'to the database'}.`
           });
         }
         
diff --git a/src/components/FocusGroupModerator.tsx b/src/components/FocusGroupModerator.tsx
index 16c85b98..964d8c9f 100644
--- a/src/components/FocusGroupModerator.tsx
+++ b/src/components/FocusGroupModerator.tsx
@@ -527,7 +527,7 @@ export default function FocusGroupModerator({ draftToEdit, onDraftSaved, preSele
       focusGroupName: "",
       discussionTopics: "",
       duration: "60",
-      llm_model: "gemini-2.5-pro",
+      llm_model: "gemini-3-pro-preview",
       reasoning_effort: "medium",
       verbosity: "medium",
     },
@@ -554,7 +554,7 @@ export default function FocusGroupModerator({ draftToEdit, onDraftSaved, preSele
         objective: values.researchBrief || '',
         topic: values.discussionTopics || '',
         duration: values.duration ? parseInt(values.duration) : 60,
-        llm_model: values.llm_model || 'gemini-2.5-pro',
+        llm_model: values.llm_model || 'gemini-3-pro-preview',
         reasoning_effort: values.reasoning_effort || 'medium',
         verbosity: values.verbosity || 'medium',
         participants: selectedParticipants,
@@ -692,7 +692,7 @@ export default function FocusGroupModerator({ draftToEdit, onDraftSaved, preSele
             topic: values.discussionTopics || '',
             description: values.researchBrief || '',
             objective: values.researchBrief || '',
-            llm_model: values.llm_model || 'gemini-2.5-pro',
+            llm_model: values.llm_model || 'gemini-3-pro-preview',
             reasoning_effort: values.reasoning_effort || 'medium',
             verbosity: values.verbosity || 'medium',
             discussionGuide: sourceFocusGroup.discussionGuide
@@ -860,7 +860,7 @@ export default function FocusGroupModerator({ draftToEdit, onDraftSaved, preSele
         objective: draftToEdit.description || draftToEdit.objective || '',
         topic: draftToEdit.topic || '',
         duration: draftToEdit.duration || 60,
-        llm_model: draftToEdit.llm_model || 'gemini-2.5-pro',
+        llm_model: draftToEdit.llm_model || 'gemini-3-pro-preview',
         reasoning_effort: draftToEdit.reasoning_effort || 'medium',
         verbosity: draftToEdit.verbosity || 'medium',
         participants: draftToEdit.participants || [],
@@ -1580,7 +1580,7 @@ true;
                             </SelectTrigger>
                           </FormControl>
                           <SelectContent>
-                            <SelectItem value="gemini-2.5-pro">Gemini 2.5 Pro</SelectItem>
+                            <SelectItem value="gemini-3-pro-preview">Gemini 3 Pro</SelectItem>
                             <SelectItem value="gpt-4.1">GPT-4.1</SelectItem>
                             <SelectItem value="gpt-5">GPT-5</SelectItem>
                           </SelectContent>
diff --git a/src/components/ai-recruiter/AIRecruiterForm.tsx b/src/components/ai-recruiter/AIRecruiterForm.tsx
index d533b73b..b2e473e7 100644
--- a/src/components/ai-recruiter/AIRecruiterForm.tsx
+++ b/src/components/ai-recruiter/AIRecruiterForm.tsx
@@ -70,7 +70,7 @@ export default function AIRecruiterForm({ onSubmit, isGenerating }: AIRecruiterF
       researchObjective: "",
       personaCount: "5",
       temperature: 1.0,
-      llm_model: "gemini-2.5-pro",
+      llm_model: "gemini-3-pro-preview",
     },
   });
 
@@ -380,7 +380,7 @@ export default function AIRecruiterForm({ onSubmit, isGenerating }: AIRecruiterF
                     </SelectTrigger>
                   </FormControl>
                   <SelectContent>
-                    <SelectItem value="gemini-2.5-pro">Gemini 2.5 Pro</SelectItem>
+                    <SelectItem value="gemini-3-pro-preview">Gemini 3 Pro</SelectItem>
                     <SelectItem value="gpt-4.1">GPT-4.1</SelectItem>
                     <SelectItem value="gpt-5">GPT-5</SelectItem>
                   </SelectContent>
diff --git a/src/components/persona/PersonaModificationModal.tsx b/src/components/persona/PersonaModificationModal.tsx
index 96899c40..bff11558 100644
--- a/src/components/persona/PersonaModificationModal.tsx
+++ b/src/components/persona/PersonaModificationModal.tsx
@@ -70,7 +70,7 @@ export default function PersonaModificationModal({
     resolver: zodResolver(modificationFormSchema),
     defaultValues: {
       modificationPrompt: "",
-      llm_model: "gemini-2.5-pro",
+      llm_model: "gemini-3-pro-preview",
       reasoning_effort: "medium",
       verbosity: "medium",
     },
@@ -221,7 +221,7 @@ export default function PersonaModificationModal({
                       </SelectTrigger>
                     </FormControl>
                     <SelectContent>
-                      <SelectItem value="gemini-2.5-pro">Gemini 2.5 Pro</SelectItem>
+                      <SelectItem value="gemini-3-pro-preview">Gemini 3 Pro</SelectItem>
                       <SelectItem value="gpt-4.1">GPT-4.1</SelectItem>
                       <SelectItem value="gpt-5">GPT-5</SelectItem>
                     </SelectContent>
diff --git a/src/pages/FocusGroupSession.tsx b/src/pages/FocusGroupSession.tsx
index de95149f..0e3e454c 100644
--- a/src/pages/FocusGroupSession.tsx
+++ b/src/pages/FocusGroupSession.tsx
@@ -730,11 +730,11 @@ const FocusGroupSession = () => {
           duration: data.duration || 60,
           topic: data.topic || 'general',
           discussionGuide: data.discussionGuide || '',
-          llm_model: data.llm_model || 'gemini-2.5-pro'
+          llm_model: data.llm_model || 'gemini-3-pro-preview'
         };
         
         setFocusGroup(focusGroupData);
-        setSelectedModel(focusGroupData.llm_model || 'gemini-2.5-pro');
+        setSelectedModel(focusGroupData.llm_model || 'gemini-3-pro-preview');
         setSelectedReasoningEffort(focusGroupData.reasoning_effort || 'medium');
         setSelectedVerbosity(focusGroupData.verbosity || 'medium');
         
@@ -816,7 +816,7 @@ const FocusGroupSession = () => {
         } : null);
         toastService.success('AI Model Updated', {
           description: `Focus group will now use ${
-            newModel === 'gemini-2.5-pro' ? 'Gemini 2.5 Pro' : 
+            newModel === 'gemini-3-pro-preview' ? 'Gemini 3 Pro' : 
             newModel === 'gpt-4.1' ? 'GPT-4.1' : 
             newModel === 'gpt-5' ? 'GPT-5' : newModel
           } for AI responses`
@@ -869,11 +869,11 @@ const FocusGroupSession = () => {
             duration: data.duration || 60,
             topic: data.topic || 'general',
             discussionGuide: data.discussionGuide || '',
-            llm_model: data.llm_model || 'gemini-2.5-pro'
+            llm_model: data.llm_model || 'gemini-3-pro-preview'
           };
           
           setFocusGroup(focusGroupData);
-          setSelectedModel(focusGroupData.llm_model || 'gemini-2.5-pro');
+          setSelectedModel(focusGroupData.llm_model || 'gemini-3-pro-preview');
         setSelectedReasoningEffort(focusGroupData.reasoning_effort || 'medium');
         setSelectedVerbosity(focusGroupData.verbosity || 'medium');
           
@@ -1932,7 +1932,7 @@ const FocusGroupSession = () => {
                 <Bot className="h-3 w-3 text-slate-500 mr-1" />
                 <Badge variant="secondary" className="text-xs">
                   {focusGroup.llm_model === 'gpt-4.1' ? 'GPT-4.1' : 
-                   focusGroup.llm_model === 'gpt-5' ? 'GPT-5' : 'Gemini 2.5 Pro'}
+                   focusGroup.llm_model === 'gpt-5' ? 'GPT-5' : 'Gemini 3 Pro'}
                 </Badge>
               </div>
             </div>
@@ -2294,7 +2294,7 @@ const FocusGroupSession = () => {
               <span className="text-sm font-medium">Current Model:</span>
               <Badge variant="secondary">
                 {focusGroup?.llm_model === 'gpt-4.1' ? 'GPT-4.1' : 
-                 focusGroup?.llm_model === 'gpt-5' ? 'GPT-5' : 'Gemini 2.5 Pro'}
+                 focusGroup?.llm_model === 'gpt-5' ? 'GPT-5' : 'Gemini 3 Pro'}
               </Badge>
             </div>
             
@@ -2308,7 +2308,7 @@ const FocusGroupSession = () => {
                   <SelectValue placeholder="Select AI model" />
                 </SelectTrigger>
                 <SelectContent>
-                  <SelectItem value="gemini-2.5-pro">Gemini 2.5 Pro</SelectItem>
+                  <SelectItem value="gemini-3-pro-preview">Gemini 3 Pro</SelectItem>
                   <SelectItem value="gpt-4.1">GPT-4.1</SelectItem>
                   <SelectItem value="gpt-5">GPT-5</SelectItem>
                 </SelectContent>
@@ -2364,7 +2364,7 @@ Controls how thoroughly GPT-5 thinks and how detailed responses are
             )}
             
             <div className="text-xs text-slate-600">
-              <p><strong>Gemini 2.5 Pro:</strong> Google's advanced model, great for creative and analytical tasks.</p>
+              <p><strong>Gemini 3 Pro:</strong> Google's advanced model, great for creative and analytical tasks.</p>
               <p><strong>GPT-4.1:</strong> OpenAI's latest model, excellent for conversational and reasoning tasks.</p>
               <p><strong>GPT-5:</strong> OpenAI's newest model with advanced reasoning and customizable response styles.</p>
             </div>
diff --git a/src/pages/SyntheticUsers.tsx b/src/pages/SyntheticUsers.tsx
index b2001e74..4ef9327a 100644
--- a/src/pages/SyntheticUsers.tsx
+++ b/src/pages/SyntheticUsers.tsx
@@ -142,7 +142,7 @@ const SyntheticUsers = () => {
   });
   // LLM selection for download
   const [downloadLlmModalOpen, setDownloadLlmModalOpen] = useState(false);
-  const [selectedDownloadLlmModel, setSelectedDownloadLlmModel] = useState<string>('gemini-2.5-pro');
+  const [selectedDownloadLlmModel, setSelectedDownloadLlmModel] = useState<string>('gemini-3-pro-preview');
   
   // Bulk export no longer needs state - direct download
   
@@ -1057,7 +1057,7 @@ const SyntheticUsers = () => {
       summaryGenerationControls.completeGeneration();
       
       // Show success toast with details including model information
-      const modelDisplayName = selectedDownloadLlmModel === 'gpt-4.1' ? 'GPT-4.1' : 'Gemini 2.5 Pro';
+      const modelDisplayName = selectedDownloadLlmModel === 'gpt-4.1' ? 'GPT-4.1' : 'Gemini 3 Pro';
       if (summary_stats.total_successful === summary_stats.total_requested) {
         toastService.success("Persona summary downloaded", {
           description: `Successfully processed all ${summary_stats.total_successful} persona${summary_stats.total_successful !== 1 ? 's' : ''} from "${folderName}" using ${modelDisplayName}`
@@ -1929,9 +1929,9 @@ const SyntheticUsers = () => {
                         className="space-y-3"
                       >
                         <div className="flex items-center space-x-2">
-                          <RadioGroupItem value="gemini-2.5-pro" id="download-gemini" />
+                          <RadioGroupItem value="gemini-3-pro-preview" id="download-gemini" />
                           <Label htmlFor="download-gemini" className="text-sm font-medium">
-                            Gemini 2.5 Pro
+                            Gemini 3 Pro
                           </Label>
                         </div>
                         <div className="flex items-center space-x-2">
diff --git a/src/utils/personaGenerator.ts b/src/utils/personaGenerator.ts
index c6564cee..0a86c3fe 100644
--- a/src/utils/personaGenerator.ts
+++ b/src/utils/personaGenerator.ts
@@ -27,7 +27,7 @@ export async function generateSyntheticPersonas(
 ): Promise<{personas: Persona[], task_id?: string, partial_success?: boolean, errors?: any[]}> {
   // Debug logging for folder and model
   console.log(`generateSyntheticPersonas called with targetFolderId: ${targetFolderId || 'none'}`);
-  console.log(`🔄 generateSyntheticPersonas using model: ${llmModel || 'gemini-2.5-pro'}`);
+  console.log(`🔄 generateSyntheticPersonas using model: ${llmModel || 'gemini-3-pro-preview'}`);
   
   try {
     // We'll use the two-stage approach