Upgrade AI models to each vendor's cheap tier and centralize model IDs behind stable vendor keys

Claude Haiku 4.5, GPT-5.6 Luna, Gemini 3.6 Flash, and Grok 4.6 everywhere
(Archive Assistant stays Sonnet, bumped to Sonnet 5). Model IDs now live in
one place per platform: utils/llm_models.py on web, AskAIModels.swift on iOS,
AskAiModels.kt on Android. Registry keys are stable vendor slugs with aliases
for legacy keys, so old clients and saved preferences keep working.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: samuelclay

apps/archive_assistant/models.py

@@ -9,6 +9,8 @@
 
 import mongoengine as mongo
 
+from utils.llm_models import ANTHROPIC_ARCHIVE_MODEL
+
 
 class MArchiveConversation(mongo.Document):
     """
@@ -54,7 +56,7 @@ class MArchiveQuery(mongo.Document):
     response_date = mongo.DateTimeField()
 
     # Metadata
-    model = mongo.StringField(max_length=64, default="claude-sonnet-4-5")
+    model = mongo.StringField(max_length=64, default=ANTHROPIC_ARCHIVE_MODEL)
     duration_ms = mongo.IntField()  # Time to generate response
     tokens_used = mongo.IntField()
 

apps/archive_assistant/tasks.py

@@ -16,15 +16,12 @@
 from django.conf import settings
 from django.contrib.auth.models import User
 
-from apps.archive_assistant.models import (
-    MArchiveAssistantUsage,
-    MArchiveConversation,
-    MArchiveQuery,
-)
+from apps.archive_assistant.models import MArchiveAssistantUsage, MArchiveConversation, MArchiveQuery
 from apps.archive_assistant.prompts import get_system_prompt
 from apps.archive_assistant.tools import ARCHIVE_TOOLS, execute_tool
 from utils import log as logging
 from utils.llm_costs import LLMCostTracker
+from utils.llm_models import ANTHROPIC_ARCHIVE_MODEL
 
 # Character limit for non-premium users before truncation
 FREE_RESPONSE_CHAR_LIMIT = 300
@@ -37,7 +34,7 @@ def get_redis_pubsub_connection():
 
 @shared_task(name="archive-assistant-query")
 def process_archive_query(
-    user_id, conversation_id, query_id, query_text, model="claude-sonnet-4-5", is_premium_archive=True
+    user_id, conversation_id, query_id, query_text, model=ANTHROPIC_ARCHIVE_MODEL, is_premium_archive=True
 ):
     """
     Process an Archive Assistant query using Claude with tools.

apps/archive_assistant/views.py

@@ -12,18 +12,15 @@
 from django.views.decorators.csrf import csrf_exempt
 from django.views.decorators.http import require_http_methods
 
-from apps.archive_assistant.models import (
-    MArchiveAssistantUsage,
-    MArchiveConversation,
-    MArchiveQuery,
-)
+from apps.archive_assistant.models import MArchiveAssistantUsage, MArchiveConversation, MArchiveQuery
 from apps.archive_assistant.prompts import get_suggested_questions
 from apps.archive_assistant.tasks import process_archive_query
 from apps.archive_extension.models import MArchivedStory
 from apps.archive_extension.utils import format_datetime_utc
 from apps.profile.models import Profile
 from utils import json_functions as json
 from utils import log as logging
+from utils.llm_models import ANTHROPIC_ARCHIVE_MODEL
 from utils.user_functions import ajax_login_required, get_user
 
 
@@ -51,7 +48,7 @@ def submit_query(request):
     POST params:
         query: The question to ask (required)
         conversation_id: Existing conversation ID (optional, creates new if not provided)
-        model: Model to use (optional, defaults to claude-sonnet-4-5)
+        model: Model to use (optional, defaults to ANTHROPIC_ARCHIVE_MODEL in utils/llm_models.py)
 
     Returns:
         {
@@ -72,7 +69,7 @@ def submit_query(request):
 
     query_text = request.POST.get("query", "").strip()
     conversation_id = request.POST.get("conversation_id", "").strip()
-    model = request.POST.get("model", "claude-sonnet-4-5")
+    model = request.POST.get("model", ANTHROPIC_ARCHIVE_MODEL)
 
     # Check if user has premium archive for full responses
     profile = Profile.objects.get(user=user)

apps/archive_extension/tasks.py

@@ -14,6 +14,7 @@
 
 from utils import json_functions as json
 from utils import log as logging
+from utils.llm_models import ANTHROPIC_MODEL
 
 
 def _publish_category_update(user_id, archive_id, categories):
@@ -212,7 +213,7 @@ def _categorize_single_archive(archive, user_categories=None):
         )
 
         response = client.messages.create(
-            model="claude-haiku-4-5",
+            model=ANTHROPIC_MODEL,
             max_tokens=100,
             messages=[{"role": "user", "content": prompt}],
         )

apps/archive_extension/views.py

@@ -15,17 +15,14 @@
 from django.views.decorators.csrf import csrf_exempt
 from django.views.decorators.http import require_http_methods
 
-from apps.archive_extension.blocklist import (
-    get_blocked_domains,
-    get_blocked_patterns,
-    is_blocked,
-)
+from apps.archive_extension.blocklist import get_blocked_domains, get_blocked_patterns, is_blocked
 from apps.archive_extension.matching import match_and_process
 from apps.archive_extension.models import MArchivedStory, MArchiveUserSettings
 from apps.archive_extension.tasks import categorize_archives, index_archive_for_search
 from apps.archive_extension.utils import format_datetime_utc
 from utils import json_functions as json
 from utils import log as logging
+from utils.llm_models import ANTHROPIC_MODEL
 from utils.user_functions import ajax_login_required, get_user
 
 
@@ -1131,7 +1128,7 @@ def split_category(request):
 ]"""
 
             response = client.messages.create(
-                model="claude-haiku-4-5", max_tokens=2000, messages=[{"role": "user", "content": prompt}]
+                model=ANTHROPIC_MODEL, max_tokens=2000, messages=[{"role": "user", "content": prompt}]
             )
 
             result_text = response.content[0].text.strip()

apps/ask_ai/models.py

@@ -13,7 +13,7 @@ class MAskAIResponse(mongo.Document):
     story_hash = mongo.StringField(max_length=32)
     question_id = mongo.StringField(max_length=64)
     custom_question = mongo.StringField()
-    model = mongo.StringField(max_length=32)  # AI model used (opus, gpt-5.2, gemini-3, grok-4.1)
+    model = mongo.StringField(max_length=32)  # AI model used (anthropic, openai, google, xai)
     response_z = mongo.BinaryField()
     response_metadata = mongo.DictField()
     created_at = mongo.DateTimeField(default=datetime.datetime.now)
@@ -56,7 +56,7 @@ def get_cached_response(cls, user_id, story_hash, question_id, custom_question=N
             story_hash: Story hash
             question_id: Question ID (e.g., "sentence", "bullets")
             custom_question: Optional custom question text
-            model: Optional model name (opus, gpt-5.2, gemini-3, grok-4.1)
+            model: Optional model name (anthropic, openai, google, xai)
 
         Returns:
             MAskAIResponse instance or None
@@ -91,7 +91,7 @@ def create_response(
             question_id: Question ID
             response_text: Full response text from AI
             custom_question: Optional custom question
-            model: Optional model name (opus, gpt-5.2, gemini-3, grok-4.1)
+            model: Optional model name (anthropic, openai, google, xai)
             metadata: Optional metadata dict (tokens, model, etc.)
 
         Returns:

apps/ask_ai/providers.py

@@ -10,6 +10,16 @@
 from google.genai import types as genai_types
 
 from utils import log as logging
+from utils.llm_models import (
+    ANTHROPIC_MODEL,
+    ANTHROPIC_MODEL_DISPLAY,
+    GOOGLE_MODEL,
+    GOOGLE_MODEL_DISPLAY,
+    OPENAI_MODEL,
+    OPENAI_MODEL_DISPLAY,
+    XAI_MODEL,
+    XAI_MODEL_DISPLAY,
+)
 
 
 def _is_placeholder(key):
@@ -372,13 +382,14 @@ def format_error(self, error: Exception) -> str:
 }
 
 # Model registry: single source of truth for all model configuration.
-# Each entry contains everything needed by both backend and frontend.
-# To add/update a model, only change this dict — frontend dropdowns are populated from it.
+# Registry keys are stable vendor slugs so model upgrades never rename keys —
+# to upgrade a model, only change the IDs/display names in utils/llm_models.py.
+# Frontend dropdowns are populated from this dict.
 _DEFAULT_MODELS = {
-    "opus": {
+    "anthropic": {
         "provider_class": AnthropicProvider,
-        "model_id": "claude-opus-4-6",
-        "display_name": "Claude Opus 4.6",
+        "model_id": ANTHROPIC_MODEL,
+        "display_name": ANTHROPIC_MODEL_DISPLAY,
         "vendor": "anthropic",
         "vendor_display": "Anthropic",
         "order": 1,
@@ -387,39 +398,61 @@ def format_error(self, error: Exception) -> str:
             "max_tokens": 16384,
         },
     },
-    "gpt-5.2": {
+    "openai": {
         "provider_class": OpenAIProvider,
-        "model_id": "gpt-5.2",
-        "display_name": "GPT 5.2",
+        "model_id": OPENAI_MODEL,
+        "display_name": OPENAI_MODEL_DISPLAY,
         "vendor": "openai",
         "vendor_display": "OpenAI",
         "order": 2,
         "thinking_config": {
             "reasoning_effort": "high",
         },
     },
-    "gemini-3": {
+    "google": {
         "provider_class": GeminiProvider,
-        "model_id": "gemini-3-pro-preview",
-        "display_name": "Gemini 3 Pro",
+        "model_id": GOOGLE_MODEL,
+        "display_name": GOOGLE_MODEL_DISPLAY,
         "vendor": "google",
         "vendor_display": "Google",
         "order": 3,
         "thinking_config": {
             "thinking_budget": -1,
         },
     },
-    "grok-4.1": {
+    "xai": {
         "provider_class": XAIProvider,
-        "model_id": "grok-4-1-fast-non-reasoning",
-        "display_name": "Grok 4.1 Fast",
+        "model_id": XAI_MODEL,
+        "display_name": XAI_MODEL_DISPLAY,
         "vendor": "xai",
         "vendor_display": "xAI",
         "order": 4,
-        "thinking_model_id": "grok-4-1-fast-reasoning",
     },
 }
 
+# Legacy registry keys from before keys became vendor slugs (Aug 2026). Old
+# mobile clients and stored user preferences still send these — resolve them
+# instead of falling back to the default model. Never reuse these as new keys.
+MODEL_ALIASES = {
+    # Legacy Ask AI keys
+    "opus": "anthropic",
+    "gpt-5.2": "openai",
+    "gemini-3": "google",
+    "grok-4.1": "xai",
+    # Legacy briefing keys
+    "haiku": "anthropic",
+    "gpt-5-mini": "openai",
+    "gemini-flash-lite": "google",
+    "grok-4.1-fast": "xai",
+}
+
+
+def _resolve_key(model_name, registry):
+    """Resolve a model key against a registry, mapping legacy keys via MODEL_ALIASES."""
+    if model_name in registry:
+        return model_name
+    return MODEL_ALIASES.get(model_name, model_name)
+
 
 def _load_models():
     """Load models, applying settings override if ASK_AI_MODELS is defined.
@@ -462,15 +495,23 @@ def _load_models():
 
 MODELS = _load_models()
 VALID_MODELS = list(MODELS.keys())
-DEFAULT_MODEL = getattr(settings, "ASK_AI_MODEL", "opus")
+DEFAULT_MODEL = _resolve_key(getattr(settings, "ASK_AI_MODEL", "anthropic"), MODELS)
 
 # MODEL_VENDORS includes both current and historical models for metrics tracking.
 # When retiring a model, remove it from MODELS above but keep it here.
 MODEL_VENDORS = {
     **{key: m["vendor"] for key, m in MODELS.items()},
     # Historical models (kept for metrics)
+    "opus": "anthropic",
+    "haiku": "anthropic",
+    "gpt-5.2": "openai",
+    "gpt-5-mini": "openai",
     "gpt-5.1": "openai",
     "gpt-4.1": "openai",
+    "gemini-3": "google",
+    "gemini-flash-lite": "google",
+    "grok-4.1": "xai",
+    "grok-4.1-fast": "xai",
     "grok-4": "xai",
 }
 
@@ -502,6 +543,7 @@ def get_provider(model_name: str, thinking: bool = False) -> tuple[LLMProvider,
     Returns:
         Tuple of (provider_instance, model_id, thinking_config)
     """
+    model_name = _resolve_key(model_name, MODELS)
     if model_name not in MODELS:
         model_name = DEFAULT_MODEL
 
@@ -516,36 +558,37 @@ def get_provider(model_name: str, thinking: bool = False) -> tuple[LLMProvider,
 
 
 # Briefing model registry: cheap models optimized for daily briefing generation.
-# Separate from the Ask AI models (which use flagship models).
+# Keys are the same stable vendor slugs as the Ask AI registry, and both tiers
+# now point at the same cheap models from utils/llm_models.py.
 _DEFAULT_BRIEFING_MODELS = {
-    "haiku": {
+    "anthropic": {
         "provider_class": AnthropicProvider,
-        "model_id": "claude-haiku-4-5",
-        "display_name": "Claude Haiku",
+        "model_id": ANTHROPIC_MODEL,
+        "display_name": ANTHROPIC_MODEL_DISPLAY,
         "vendor": "anthropic",
         "vendor_display": "Anthropic",
         "order": 1,
     },
-    "gpt-5-mini": {
+    "openai": {
         "provider_class": OpenAIProvider,
-        "model_id": "gpt-5-mini",
-        "display_name": "GPT 5 Mini",
+        "model_id": OPENAI_MODEL,
+        "display_name": OPENAI_MODEL_DISPLAY,
         "vendor": "openai",
         "vendor_display": "OpenAI",
         "order": 2,
     },
-    "gemini-flash-lite": {
+    "google": {
         "provider_class": GeminiProvider,
-        "model_id": "gemini-2.5-flash-lite",
-        "display_name": "Gemini Flash Lite",
+        "model_id": GOOGLE_MODEL,
+        "display_name": GOOGLE_MODEL_DISPLAY,
         "vendor": "google",
         "vendor_display": "Google",
         "order": 3,
     },
-    "grok-4.1-fast": {
+    "xai": {
         "provider_class": XAIProvider,
-        "model_id": "grok-4-1-fast-non-reasoning",
-        "display_name": "Grok 4.1 Fast",
+        "model_id": XAI_MODEL,
+        "display_name": XAI_MODEL_DISPLAY,
         "vendor": "xai",
         "vendor_display": "xAI",
         "order": 4,
@@ -589,7 +632,19 @@ def _load_briefing_models():
 
 BRIEFING_MODELS = _load_briefing_models()
 VALID_BRIEFING_MODELS = list(BRIEFING_MODELS.keys())
-DEFAULT_BRIEFING_MODEL = getattr(settings, "BRIEFING_MODEL", "haiku")
+DEFAULT_BRIEFING_MODEL = _resolve_key(getattr(settings, "BRIEFING_MODEL", "anthropic"), BRIEFING_MODELS)
+
+
+def resolve_briefing_model_key(model_name: Optional[str]) -> Optional[str]:
+    """Resolve a stored/POSTed briefing model key to a current registry key.
+
+    Maps legacy keys ("haiku") to vendor keys ("anthropic"). Returns None when
+    the key isn't recognized so callers can fall back to the server default.
+    """
+    if not model_name:
+        return None
+    model_name = _resolve_key(model_name, BRIEFING_MODELS)
+    return model_name if model_name in BRIEFING_MODELS else None
 
 
 def get_briefing_models_for_frontend() -> list:
@@ -613,7 +668,6 @@ def get_briefing_models_for_frontend() -> list:
 
 def get_briefing_provider(model_name: str) -> tuple[LLMProvider, str]:
     """Get a provider instance and model ID for the given briefing model name."""
-    if not model_name or model_name not in BRIEFING_MODELS:
-        model_name = DEFAULT_BRIEFING_MODEL
+    model_name = resolve_briefing_model_key(model_name) or DEFAULT_BRIEFING_MODEL
     model = BRIEFING_MODELS[model_name]
     return model["provider_class"](), model["model_id"]