Skip to content

OpenRAG is a comprehensive, single package Retrieval-Augmented Generation platform built on Langflow, Docling, and Opensearch.

Python

feat: make functional and up to date models available in the list (#2164)

Wallgau591c03407403
diff --git a/frontend/app/settings/_helpers/model-helpers.tsx b/frontend/app/settings/_helpers/model-helpers.tsx
index fac3d2ee0..e658a3888 100644
--- a/frontend/app/settings/_helpers/model-helpers.tsx
+++ b/frontend/app/settings/_helpers/model-helpers.tsx
@@ -94,14 +94,26 @@ export function getModelLogo(modelValue: string, provider?: ModelProvider) {
return <OpenAILogo className="w-4 h-4" />; // Default to OpenAI logo
}
-// Helper function to get fallback models by provider
+// Offline fallbacks when live /api/models/* returns empty.
+// Include current preferred defaults AND older models that are still functional.
+// Live provider lists remain the real catalog when the API is reachable.
export function getFallbackModels(provider: ModelProvider) {
switch (provider) {
case "openai":
return {
language: [
+ // GPT-5.6 family (current frontier)
+ { value: "gpt-5.6", label: "GPT-5.6" },
+ { value: "gpt-5.6-sol", label: "GPT-5.6 Sol" },
+ { value: "gpt-5.6-terra", label: "GPT-5.6 Terra" },
+ { value: "gpt-5.6-luna", label: "GPT-5.6 Luna" },
+ // GPT-5.5
+ { value: "gpt-5.5", label: "GPT-5.5" },
+ { value: "gpt-5.5-pro", label: "GPT-5.5 Pro" },
+ // GPT-5.4 and earlier (still functional)
{ value: "gpt-5.4", label: "GPT-5.4" },
{ value: "gpt-5.4-mini", label: "GPT-5.4 Mini" },
+ { value: "gpt-5.4-nano", label: "GPT-5.4 Nano" },
{ value: "gpt-5.4-pro", label: "GPT-5.4 Pro" },
{ value: "gpt-5.3-codex", label: "GPT-5.3 Codex" },
{ value: "gpt-5.2", label: "GPT-5.2" },
@@ -109,78 +121,64 @@ export function getFallbackModels(provider: ModelProvider) {
{ value: "gpt-5", label: "GPT-5" },
{ value: "gpt-5-mini", label: "GPT-5 Mini" },
{ value: "gpt-5-nano", label: "GPT-5 Nano" },
- { value: "gpt-4o-mini", label: "GPT-4o Mini" },
- { value: "gpt-4o", label: "GPT-4o" },
{ value: "gpt-4.1", label: "GPT-4.1" },
{ value: "gpt-4.1-mini", label: "GPT-4.1 Mini" },
+ { value: "gpt-4o", label: "GPT-4o" },
+ { value: "gpt-4o-mini", label: "GPT-4o Mini" },
{ value: "o3", label: "o3" },
{ value: "o3-pro", label: "o3 Pro" },
+ { value: "o4-mini", label: "o4 Mini" },
{ value: "o4-mini-high", label: "o4 Mini High" },
],
embedding: [
- { value: "text-embedding-ada-002", label: "text-embedding-ada-002" },
- { value: "text-embedding-3-small", label: "text-embedding-3-small" },
{ value: "text-embedding-3-large", label: "text-embedding-3-large" },
+ { value: "text-embedding-3-small", label: "text-embedding-3-small" },
+ { value: "text-embedding-ada-002", label: "text-embedding-ada-002" },
],
};
case "anthropic":
return {
language: [
+ // Claude 5 family (current)
+ { value: "claude-fable-5", label: "Claude Fable 5" },
+ { value: "claude-opus-5", label: "Claude Opus 5" },
+ { value: "claude-sonnet-5", label: "Claude Sonnet 5" },
+ // Claude 4.x (still functional)
+ { value: "claude-opus-4-8", label: "Claude Opus 4.8" },
+ { value: "claude-opus-4-7", label: "Claude Opus 4.7" },
{ value: "claude-opus-4-6", label: "Claude Opus 4.6" },
{ value: "claude-sonnet-4-6", label: "Claude Sonnet 4.6" },
+ { value: "claude-opus-4-5-20251101", label: "Claude Opus 4.5" },
{ value: "claude-sonnet-4-5-20250929", label: "Claude Sonnet 4.5" },
+ { value: "claude-haiku-4-5-20251001", label: "Claude Haiku 4.5" },
],
};
case "ollama":
return {
+ // Tool-calling capable recommendations only (agent requires tools)
language: [
- { value: "llama2", label: "Llama 2" },
- { value: "llama2:13b", label: "Llama 2 13B" },
- { value: "codellama", label: "Code Llama" },
+ { value: "gpt-oss", label: "gpt-oss" },
+ { value: "mistral-nemo", label: "mistral-nemo" },
+ { value: "llama3.1", label: "Llama 3.1" },
+ { value: "qwen2.5", label: "Qwen 2.5" },
],
embedding: [
- { value: "mxbai-embed-large", label: "MxBai Embed Large" },
{ value: "nomic-embed-text", label: "Nomic Embed Text" },
+ { value: "mxbai-embed-large", label: "MxBai Embed Large" },
],
};
case "watsonx":
- return {
- language: [
- {
- value: "meta-llama/llama-3-1-70b-instruct",
- label: "Llama 3.1 70B Instruct",
- },
- { value: "ibm/granite-13b-chat-v2", label: "Granite 13B Chat v2" },
- ],
- embedding: [
- {
- value: "ibm/slate-125m-english-rtrvr",
- label: "Slate 125M English Retriever",
- },
- ],
- };
+ // No stable static IDs — live list is required for watsonx.
+ return { language: [], embedding: [] };
default:
return {
language: [
- { value: "gpt-5.4", label: "GPT-5.4" },
+ { value: "gpt-5.6-luna", label: "GPT-5.6 Luna" },
{ value: "gpt-5.4-mini", label: "GPT-5.4 Mini" },
- { value: "gpt-5.4-pro", label: "GPT-5.4 Pro" },
- { value: "gpt-5.3-codex", label: "GPT-5.3 Codex" },
- { value: "gpt-5.2", label: "GPT-5.2" },
- { value: "gpt-5.1", label: "GPT-5.1" },
- { value: "gpt-5", label: "GPT-5" },
- { value: "gpt-5-mini", label: "GPT-5 Mini" },
- { value: "gpt-5-nano", label: "GPT-5 Nano" },
- { value: "gpt-4o-mini", label: "GPT-4o Mini" },
{ value: "gpt-4o", label: "GPT-4o" },
- { value: "gpt-4.1", label: "GPT-4.1" },
- { value: "gpt-4.1-mini", label: "GPT-4.1 Mini" },
- { value: "o3", label: "o3" },
- { value: "o3-pro", label: "o3 Pro" },
- { value: "o4-mini-high", label: "o4 Mini High" },
+ { value: "gpt-4o-mini", label: "GPT-4o Mini" },
],
embedding: [
- { value: "text-embedding-ada-002", label: "text-embedding-ada-002" },
{ value: "text-embedding-3-small", label: "text-embedding-3-small" },
{ value: "text-embedding-3-large", label: "text-embedding-3-large" },
],
diff --git a/src/api/settings/helpers.py b/src/api/settings/helpers.py
index c0c7b5440..b932f7c57 100644
--- a/src/api/settings/helpers.py
+++ b/src/api/settings/helpers.py
@@ -47,7 +47,12 @@ def _first_configured_embedding_provider(config, excluding: str) -> str:
def _default_llm_model(provider: str) -> str:
- """Return the default LLM model for a provider."""
+ """Return the static preferred LLM model for a provider.
+
+ OpenAI/Anthropic have stable preferred defaults. Ollama/watsonx model IDs are
+ dynamic from the live provider list — return empty so the frontend picks from
+ the live list (default flag or first available).
+ """
from config.model_constants import (
ANTHROPIC_DEFAULT_LANGUAGE_MODEL,
OPENAI_DEFAULT_LANGUAGE_MODEL,
@@ -56,8 +61,8 @@ def _default_llm_model(provider: str) -> str:
return {
"openai": OPENAI_DEFAULT_LANGUAGE_MODEL,
"anthropic": ANTHROPIC_DEFAULT_LANGUAGE_MODEL,
- "watsonx": "ibm/granite-4-h-small",
- "ollama": "llama3",
+ "watsonx": "",
+ "ollama": "",
}.get(provider, "")
diff --git a/src/config/model_constants.py b/src/config/model_constants.py
index 401513468..22757bcef 100644
--- a/src/config/model_constants.py
+++ b/src/config/model_constants.py
@@ -1,31 +1,9 @@
-"""Shared model constants used across provider/model validation flows."""
+"""Preferred default model IDs used when a live provider list is available.
-ANTHROPIC_VALIDATION_MODELS = [
- "claude-opus-4-6",
- "claude-sonnet-4-6",
- "claude-haiku-4-5-20251001",
- "claude-opus-4-5-20251101",
- "claude-sonnet-4-5-20250929",
-]
-
-OPENAI_VALIDATION_MODELS = [
- "gpt-5.4",
- "gpt-5.4-mini",
- "gpt-5.4-pro",
- "gpt-5.3-codex",
- "gpt-5.2",
- "gpt-5.1",
- "gpt-5",
- "gpt-5-mini",
- "gpt-5-nano",
- "gpt-4o-mini",
- "gpt-4o",
- "gpt-4.1",
- "gpt-4.1-mini",
- "o3",
- "o3-pro",
- "o4-mini-high",
-]
+Live provider APIs are the source of truth for which models appear in pickers.
+These constants are soft preferences: used when present in the live list, and
+as thin offline fallbacks when a live fetch is unavailable.
+"""
OPENAI_DEFAULT_LANGUAGE_MODEL = "gpt-5.4-mini"
diff --git a/src/services/models_service.py b/src/services/models_service.py
index 078606f71..d78cdb345 100644
--- a/src/services/models_service.py
+++ b/src/services/models_service.py
@@ -1,4 +1,5 @@
import asyncio
+import re
import httpx
@@ -16,6 +17,75 @@ logger = get_logger(__name__)
KNOWN_PREFIXES = ["openai", "ollama", "watsonx", "anthropic"]
+# OpenAI /v1/models is a flat inventory. These IDs are real products but not
+# usable as OpenRAG agent LLMs (wrong modality / API surface).
+_OPENAI_NON_CHAT_PREFIXES = (
+ "whisper",
+ "dall-e",
+ "tts-",
+ "davinci",
+ "babbage",
+ "curie",
+ "sora-",
+ "computer-use",
+ "omni-moderation",
+ "text-moderation",
+ "gpt-image",
+ "gpt-audio",
+ "gpt-realtime",
+ "chatgpt-image",
+)
+_OPENAI_REASONING_MODEL_RE = re.compile(r"^o\d")
+
+
+def is_openai_embedding_model(model_id: str) -> bool:
+ """True if the OpenAI model ID is an embedding model."""
+ return OPENAI_EMBEDDING_MODEL_PREFIX in model_id or "text-similarity-" in model_id
+
+
+def is_openai_non_chat_model(model_id: str) -> bool:
+ """True if the ID is a known non-chat OpenAI product (junk for LLM/embedding pickers)."""
+ lower = model_id.lower()
+ if "-moderation" in lower:
+ return True
+ # Mid-string modality markers (e.g. gpt-4o-realtime-preview, gpt-4o-mini-tts)
+ # that prefix checks alone miss.
+ if any(marker in lower for marker in ("-realtime", "-transcribe", "-tts")):
+ return True
+ return any(lower.startswith(prefix) for prefix in _OPENAI_NON_CHAT_PREFIXES)
+
+
+def is_openai_language_model(model_id: str) -> bool:
+ """True if the OpenAI model ID is usable as a chat/agent language model.
+
+ Classifies the live /v1/models inventory: embeddings and non-chat junk are
+ excluded; gpt-*, chatgpt-*, ft:gpt-* fine-tunes, and o<digit>* reasoning
+ models are included so new chat families appear without a curated allowlist.
+ """
+ if not model_id or is_openai_embedding_model(model_id) or is_openai_non_chat_model(model_id):
+ return False
+ if model_id.startswith(("gpt-", "chatgpt-", "ft:gpt-")):
+ return True
+ return bool(_OPENAI_REASONING_MODEL_RE.match(model_id))
+
+
+def resolve_preferred_model(preferred: str, live_models: list[dict]) -> str:
+ """Pick a model from a live provider list.
+
+ Prefer ``preferred`` when it appears in the live list; otherwise the entry
+ marked ``default``, otherwise the first live model. If the live list is
+ empty, return ``preferred`` (may be empty).
+ """
+ if not live_models:
+ return preferred or ""
+ values = {m.get("value") for m in live_models}
+ if preferred and preferred in values:
+ return preferred
+ for model in live_models:
+ if model.get("default"):
+ return model.get("value") or preferred or ""
+ return live_models[0].get("value") or preferred or ""
+
class UnknownEmbeddingProvider(Exception):
"""Raised when a model's provider can't be resolved and the caller asked
@@ -230,36 +300,45 @@ class ModelsService:
data = response.json()
models = data.get("data", [])
- # Filter for relevant models
+ # Classify live inventory into embedding vs chat; drop non-chat junk.
language_models = []
embedding_models = []
for model in models:
model_id = model.get("id", "")
+ if not model_id:
+ continue
- # Embedding models
- if OPENAI_EMBEDDING_MODEL_PREFIX in model_id or "text-similarity-" in model_id:
+ if is_openai_embedding_model(model_id):
embedding_models.append(
{
"value": model_id,
"label": model_id,
- "default": model_id == OPENAI_DEFAULT_EMBEDDING_MODEL,
+ "default": False,
}
)
- # Language models (GPT and o1/o3/chatgpt models)
- elif (
- model_id.startswith(("gpt-", "o1-", "o3-", "chatgpt-"))
- and "-moderation" not in model_id
- ):
+ elif is_openai_language_model(model_id):
language_models.append(
{
"value": model_id,
"label": model_id,
- "default": model_id == OPENAI_DEFAULT_LANGUAGE_MODEL,
+ "default": False,
"supports_images": self._openai_supports_images(model_id),
}
)
+ chosen_language = resolve_preferred_model(
+ OPENAI_DEFAULT_LANGUAGE_MODEL, language_models
+ )
+ for entry in language_models:
+ entry["default"] = entry["value"] == chosen_language
+
+ chosen_embedding = resolve_preferred_model(
+ OPENAI_DEFAULT_EMBEDDING_MODEL, embedding_models
+ )
+ for entry in embedding_models:
+ entry["default"] = entry["value"] == chosen_embedding
+
# Sort by name and ensure defaults are first
language_models.sort(key=lambda x: (not x.get("default", False), x["value"]))
embedding_models.sort(key=lambda x: (not x.get("default", False), x["value"]))
@@ -302,7 +381,7 @@ class ModelsService:
"Content-Type": "application/json",
}
- # Validate API key with lightweight models endpoint and return curated models
+ # Validate API key and return all models from Anthropic's chat-oriented list API
async with httpx.AsyncClient() as client:
response = await client.get(
"https://api.anthropic.com/v1/models",
@@ -314,20 +393,28 @@ class ModelsService:
data = response.json()
models = data.get("data", [])
- # Filter for curated Anthropic models (same pattern as OpenAI validation list)
+ # Anthropic /v1/models is already chat-oriented; pass through live list.
language_models = []
for model in models:
model_id = model.get("id", "")
+ if not model_id:
+ continue
language_models.append(
{
"value": model_id,
"label": model.get("display_name", model_id),
- "default": model_id == ANTHROPIC_DEFAULT_LANGUAGE_MODEL,
+ "default": False,
"supports_images": self._anthropic_supports_images(model),
}
)
+ chosen_language = resolve_preferred_model(
+ ANTHROPIC_DEFAULT_LANGUAGE_MODEL, language_models
+ )
+ for entry in language_models:
+ entry["default"] = entry["value"] == chosen_language
+
# Sort by default first, then by name
language_models.sort(key=lambda x: (not x.get("default", False), x["value"]))
diff --git a/tests/unit/services/test_models_service_classification.py b/tests/unit/services/test_models_service_classification.py
new file mode 100644
index 000000000..6a5281188
--- /dev/null
+++ b/tests/unit/services/test_models_service_classification.py
@@ -0,0 +1,124 @@
+"""Unit tests for OpenAI model classification and preferred-model resolution."""
+
+from config.embedding_constants import OPENAI_DEFAULT_EMBEDDING_MODEL
+from config.model_constants import OPENAI_DEFAULT_LANGUAGE_MODEL
+from services.models_service import (
+ is_openai_embedding_model,
+ is_openai_language_model,
+ is_openai_non_chat_model,
+ resolve_preferred_model,
+)
+
+
+class TestOpenAIEmbeddingClassification:
+ def test_text_embedding_models(self):
+ assert is_openai_embedding_model("text-embedding-3-small")
+ assert is_openai_embedding_model("text-embedding-3-large")
+ assert is_openai_embedding_model("text-embedding-ada-002")
+
+ def test_text_similarity_models(self):
+ assert is_openai_embedding_model("text-similarity-davinci-001")
+
+ def test_chat_models_are_not_embeddings(self):
+ assert not is_openai_embedding_model("gpt-5.4-mini")
+ assert not is_openai_embedding_model("o4-mini")
+
+
+class TestOpenAINonChatClassification:
+ def test_whisper_dalle_tts_moderation(self):
+ assert is_openai_non_chat_model("whisper-1")
+ assert is_openai_non_chat_model("dall-e-3")
+ assert is_openai_non_chat_model("tts-1")
+ assert is_openai_non_chat_model("tts-1-hd")
+ assert is_openai_non_chat_model("omni-moderation-latest")
+ assert is_openai_non_chat_model("text-moderation-latest")
+ assert is_openai_non_chat_model("gpt-4o-moderation")
+
+ def test_image_audio_realtime_variants(self):
+ assert is_openai_non_chat_model("gpt-image-1")
+ assert is_openai_non_chat_model("gpt-audio-1")
+ assert is_openai_non_chat_model("gpt-realtime")
+ assert is_openai_non_chat_model("gpt-4o-realtime-preview")
+ assert is_openai_non_chat_model("gpt-4o-transcribe")
+ assert is_openai_non_chat_model("gpt-4o-mini-tts")
+
+ def test_chat_models_are_not_junk(self):
+ assert not is_openai_non_chat_model("gpt-5.4-mini")
+ assert not is_openai_non_chat_model("o4-mini")
+ assert not is_openai_non_chat_model("chatgpt-4o-latest")
+
+
+class TestOpenAILanguageClassification:
+ def test_gpt_and_chatgpt_families(self):
+ assert is_openai_language_model("gpt-5.4-mini")
+ assert is_openai_language_model("gpt-4o")
+ assert is_openai_language_model("gpt-4.1")
+ assert is_openai_language_model("chatgpt-4o-latest")
+
+ def test_reasoning_o_families_including_o4(self):
+ assert is_openai_language_model("o1")
+ assert is_openai_language_model("o1-mini")
+ assert is_openai_language_model("o3")
+ assert is_openai_language_model("o3-pro")
+ assert is_openai_language_model("o4-mini")
+ assert is_openai_language_model("o4-mini-high")
+
+ def test_excludes_embeddings_and_junk(self):
+ assert not is_openai_language_model("text-embedding-3-small")
+ assert not is_openai_language_model("whisper-1")
+ assert not is_openai_language_model("dall-e-3")
+ assert not is_openai_language_model("gpt-image-1")
+ assert not is_openai_language_model("omni-moderation-latest")
+ assert not is_openai_language_model("gpt-4o-realtime-preview")
+ assert not is_openai_language_model("gpt-4o-transcribe")
+ assert not is_openai_language_model("gpt-4o-mini-tts")
+
+ def test_excludes_empty_id(self):
+ assert not is_openai_language_model("")
+
+ def test_fine_tuned_gpt_models(self):
+ assert is_openai_language_model("ft:gpt-4o:org:custom")
+ assert is_openai_language_model("ft:gpt-4.1-mini:acme:my-tune:abc123")
+
+
+class TestResolvePreferredModel:
+ def test_uses_preferred_when_present(self):
+ live = [
+ {"value": "gpt-4o", "default": False},
+ {"value": OPENAI_DEFAULT_LANGUAGE_MODEL, "default": False},
+ ]
+ assert (
+ resolve_preferred_model(OPENAI_DEFAULT_LANGUAGE_MODEL, live)
+ == OPENAI_DEFAULT_LANGUAGE_MODEL
+ )
+
+ def test_uses_live_default_when_preferred_missing(self):
+ live = [
+ {"value": "gpt-4o", "default": False},
+ {"value": "o4-mini", "default": True},
+ ]
+ assert resolve_preferred_model("gpt-missing", live) == "o4-mini"
+
+ def test_uses_first_when_preferred_and_default_missing(self):
+ live = [
+ {"value": "o4-mini", "default": False},
+ {"value": "gpt-4o", "default": False},
+ ]
+ assert resolve_preferred_model("gpt-missing", live) == "o4-mini"
+
+ def test_empty_live_list_returns_preferred(self):
+ assert (
+ resolve_preferred_model(OPENAI_DEFAULT_LANGUAGE_MODEL, [])
+ == OPENAI_DEFAULT_LANGUAGE_MODEL
+ )
+ assert resolve_preferred_model("", []) == ""
+
+ def test_embedding_preferred(self):
+ live = [
+ {"value": "text-embedding-3-large", "default": False},
+ {"value": OPENAI_DEFAULT_EMBEDDING_MODEL, "default": False},
+ ]
+ assert (
+ resolve_preferred_model(OPENAI_DEFAULT_EMBEDDING_MODEL, live)
+ == OPENAI_DEFAULT_EMBEDDING_MODEL
+ )