Skip to content

The agent engineering platform.

feat(model-profiles): support more providers (#33766)

ccurme424214041ecc
diff --git a/libs/model-profiles/langchain_model_profiles/data/augmentations/providers/amazon-bedrock.toml b/libs/model-profiles/langchain_model_profiles/data/augmentations/providers/amazon-bedrock.toml
new file mode 100644
index 0000000000..658df1dafd
--- /dev/null
+++ b/libs/model-profiles/langchain_model_profiles/data/augmentations/providers/amazon-bedrock.toml
@@ -0,0 +1,6 @@
+[profile]
+image_url_inputs = true
+pdf_inputs = true
+pdf_tool_message = true
+image_tool_message = true
+structured_output = false
diff --git a/libs/model-profiles/langchain_model_profiles/data/augmentations/providers/google.toml b/libs/model-profiles/langchain_model_profiles/data/augmentations/providers/google.toml
new file mode 100644
index 0000000000..3b652c5d00
--- /dev/null
+++ b/libs/model-profiles/langchain_model_profiles/data/augmentations/providers/google.toml
@@ -0,0 +1,6 @@
+[profile]
+image_url_inputs = true
+pdf_inputs = true
+image_tool_message = true
+tool_choice = true
+structured_output = true
diff --git a/libs/model-profiles/langchain_model_profiles/data/augmentations/providers/google_vertexai.toml b/libs/model-profiles/langchain_model_profiles/data/augmentations/providers/google_vertexai.toml
new file mode 100644
index 0000000000..3b652c5d00
--- /dev/null
+++ b/libs/model-profiles/langchain_model_profiles/data/augmentations/providers/google_vertexai.toml
@@ -0,0 +1,6 @@
+[profile]
+image_url_inputs = true
+pdf_inputs = true
+image_tool_message = true
+tool_choice = true
+structured_output = true
diff --git a/libs/model-profiles/langchain_model_profiles/model_profile.py b/libs/model-profiles/langchain_model_profiles/model_profile.py
index b67ddb80c9..6eb06a860e 100644
--- a/libs/model-profiles/langchain_model_profiles/model_profile.py
+++ b/libs/model-profiles/langchain_model_profiles/model_profile.py
@@ -1,5 +1,7 @@
"""Model profiles package."""
+import re
+
from typing_extensions import TypedDict
from langchain_model_profiles._data_loader import _DataLoader
@@ -79,27 +81,67 @@ _LOADER = _DataLoader()
_lc_type_to_provider_id = {
"openai-chat": "openai",
+ "azure-openai-chat": "azure",
"anthropic-chat": "anthropic",
+ "chat-google-generative-ai": "google",
+ "vertexai": "google-vertex",
+ "anthropic-chat-vertexai": "google-vertex-anthropic",
+ "amazon_bedrock_chat": "amazon-bedrock",
+ "amazon_bedrock_converse_chat": "amazon-bedrock",
+ "chat-ai21": "ai21",
+ "chat-deepseek": "deepseek",
+ "fireworks-chat": "fireworks-ai",
+ "groq-chat": "groq",
+ "huggingface-chat-wrapper": "huggingface",
+ "mistralai-chat": "mistral",
+ "chat-ollama": "ollama",
+ "perplexitychat": "perplexity",
+ "together-chat": "togetherai",
+ "upstage-chat": "upstage",
+ "xai-chat": "xai",
}
-def get_model_profile(provider_id: str, model_id: str) -> ModelProfile | None:
+def _translate_provider_and_model_id(provider: str, model: str) -> tuple[str, str]:
+ """Translate LangChain provider and model to models.dev equivalents.
+
+ Args:
+ provider: LangChain provider ID.
+ model: LangChain model ID.
+
+ Returns:
+ A tuple containing the models.dev provider ID and model ID.
+ """
+ provider_id = _lc_type_to_provider_id.get(provider, provider)
+
+ if provider_id in ("google", "google-vertex"):
+ # convert models/gemini-2.0-flash-001 to gemini-2.0-flash
+ model_id = re.sub(r"-\d{3}$", "", model.replace("models/", ""))
+ elif provider_id == "amazon-bedrock":
+ # strip region prefixes like "us."
+ model_id = re.sub(r"^[A-Za-z]{2}\.", "", model)
+ else:
+ model_id = model
+
+ return provider_id, model_id
+
+
+def get_model_profile(provider: str, model: str) -> ModelProfile | None:
"""Get the model capabilities for a given model.
Args:
- provider_id: Identifier for provider (e.g., `'openai'`, `'anthropic'`).
- model_id: Identifier for model (e.g., `'gpt-5'`,
+ provider: Identifier for provider (e.g., `'openai'`, `'anthropic'`).
+ model: Identifier for model (e.g., `'gpt-5'`,
`'claude-sonnet-4-5-20250929'`).
Returns:
The model capabilities or `None` if not found in the data.
"""
- if not provider_id or not model_id:
+ if not provider or not model:
return None
- data = _LOADER.get_profile_data(
- _lc_type_to_provider_id.get(provider_id, provider_id), model_id
- )
+ provider_id, model_id = _translate_provider_and_model_id(provider, model)
+ data = _LOADER.get_profile_data(provider_id, model_id)
if not data:
# If either (1) provider not found or (2) model not found under matched provider
return None
diff --git a/libs/model-profiles/tests/unit_tests/test_chat_model.py b/libs/model-profiles/tests/unit_tests/test_chat_model.py
index ef808e5a1b..26fb9bef62 100644
--- a/libs/model-profiles/tests/unit_tests/test_chat_model.py
+++ b/libs/model-profiles/tests/unit_tests/test_chat_model.py
@@ -1,6 +1,5 @@
"""End to end test for fetching model profiles from a chat model."""
-import pytest
from langchain.chat_models import init_chat_model
@@ -12,7 +11,7 @@ def test_chat_model() -> None:
assert model.profile["structured_output"]
-def test_chat_model_no_data(monkeypatch: pytest.MonkeyPatch) -> None:
+def test_chat_model_no_data() -> None:
"""Test that chat model handles missing profile data."""
model = init_chat_model("openai:gpt-fake", api_key="foo")
assert model.profile == {}
diff --git a/libs/model-profiles/tests/unit_tests/test_model_profile.py b/libs/model-profiles/tests/unit_tests/test_model_profile.py
new file mode 100644
index 0000000000..44c40c640a
--- /dev/null
+++ b/libs/model-profiles/tests/unit_tests/test_model_profile.py
@@ -0,0 +1,31 @@
+"""Test provider and model ID mappings."""
+
+import pytest
+
+from langchain_model_profiles.model_profile import get_model_profile
+
+
+ ("provider", "model_id"),
+ [
+ ("openai-chat", "gpt-5"),
+ ("azure-openai-chat", "gpt-5"),
+ ("anthropic-chat", "claude-sonnet-4-5"),
+ ("vertexai", "models/gemini-2.0-flash-001"),
+ ("chat-google-generative-ai", "models/gemini-2.0-flash-001"),
+ ("amazon_bedrock_chat", "anthropic.claude-sonnet-4-20250514-v1:0"),
+ ("amazon_bedrock_converse_chat", "anthropic.claude-sonnet-4-20250514-v1:0"),
+ # ("chat-ai21", "jamba-mini"), # no data yet # noqa: ERA001
+ ("chat-deepseek", "deepseek-reasoner"),
+ ("fireworks-chat", "accounts/fireworks/models/gpt-oss-20b"),
+ ("groq-chat", "llama-3.3-70b-versatile"),
+ ("huggingface-chat-wrapper", "Qwen/Qwen3-235B-A22B-Thinking-2507"),
+ ("mistralai-chat", "mistral-large-latest"),
+ # ("chat-ollama", "llama3.1"), # no data yet # noqa: ERA001
+ ("perplexitychat", "sonar"),
+ ("xai-chat", "grok-4"),
+ ],
+)
+def test_id_translation(provider: str, model_id: str) -> None:
+ """Test translation from LangChain to model / provider IDs."""
+ assert get_model_profile(provider, model_id)