Skip to content

LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows

[MISC] Fix zero cost tracking for OpenAI-compatible adapter (#1985)

Chandrasekharan Mf08cd08334e0
diff --git a/unstract/sdk1/src/unstract/sdk1/adapters/base1.py b/unstract/sdk1/src/unstract/sdk1/adapters/base1.py
index c78fe5c43..467b1ff25 100644
--- a/unstract/sdk1/src/unstract/sdk1/adapters/base1.py
+++ b/unstract/sdk1/src/unstract/sdk1/adapters/base1.py
@@ -6,6 +6,7 @@ import re
from abc import ABC, abstractmethod
from importlib import import_module
from typing import TYPE_CHECKING
+from urllib.parse import urlparse
if TYPE_CHECKING:
from typing import Any
@@ -372,6 +373,16 @@ def _is_openai_reasoning_model(model: str) -> bool:
return bool(_OPENAI_REASONING_MODEL_PATTERN.match(model.lower()))
+def _is_openai_api_endpoint(api_base: str | None) -> bool:
+ """Return True when api_base points at OpenAI's own HTTPS API host."""
+ if not api_base:
+ return False
+ parsed = urlparse(api_base)
+ return (
+ parsed.scheme == "https" and (parsed.hostname or "").lower() == "api.openai.com"
+ )
+
+
class OpenAICompatibleLLMParameters(BaseChatCompletionParameters):
"""See https://docs.litellm.ai/docs/providers/openai_compatible/."""
@@ -463,6 +474,14 @@ class OpenAICompatibleLLMParameters(BaseChatCompletionParameters):
else:
validated.pop("reasoning_effort", None)
+ # The custom_openai/ prefix has no entry in the cost price map. Strip
+ # it only for OpenAI's own endpoint; other gateways price the same
+ # model name differently, so leave their cost unresolved.
+ if _is_openai_api_endpoint(validated.get("api_base")):
+ validated["cost_model"] = validated["model"][
+ len(_CUSTOM_OPENAI_PROVIDER_PREFIX) :
+ ]
+
return validated
@staticmethod
diff --git a/unstract/sdk1/tests/test_openai_compatible_adapter.py b/unstract/sdk1/tests/test_openai_compatible_adapter.py
index fb7da3165..d1dd631d2 100644
--- a/unstract/sdk1/tests/test_openai_compatible_adapter.py
+++ b/unstract/sdk1/tests/test_openai_compatible_adapter.py
@@ -303,6 +303,56 @@ def test_openai_compatible_validate_no_reasoning_unchanged() -> None:
assert "reasoning_effort" not in validated
+def test_openai_compatible_validate_sets_cost_model_for_openai_endpoint() -> None:
+ # On OpenAI's endpoint, cost_model drops the prefix so pricing resolves.
+ validated = OpenAICompatibleLLMParameters.validate(
+ {
+ "api_base": "https://api.openai.com/v1",
+ "api_key": "test-key",
+ "model": "gpt-4o",
+ }
+ )
+
+ assert validated["model"] == "custom_openai/gpt-4o"
+ assert validated["cost_model"] == "gpt-4o"
+
+
+def test_openai_compatible_validate_cost_model_keeps_openai_subprefix() -> None:
+ validated = OpenAICompatibleLLMParameters.validate(
+ {
+ "api_base": "https://api.openai.com/v1",
+ "model": "custom_openai/openai/gpt-4o",
+ }
+ )
+
+ assert validated["cost_model"] == "openai/gpt-4o"
+
+
+def test_openai_compatible_validate_no_cost_model_for_other_gateway() -> None:
+ # Non-OpenAI gateways price the same name differently; leave it unresolved.
+ validated = OpenAICompatibleLLMParameters.validate(
+ {
+ "api_base": "https://gateway.example.com/v1",
+ "model": "gpt-4o",
+ }
+ )
+
+ assert "cost_model" not in validated
+
+
+def test_openai_compatible_validate_cost_model_stable_on_revalidation() -> None:
+ # validate() may run on its own previous output; cost_model must survive.
+ first = OpenAICompatibleLLMParameters.validate(
+ {
+ "api_base": "https://api.openai.com/v1",
+ "model": "gpt-4o",
+ }
+ )
+ second = OpenAICompatibleLLMParameters.validate(first)
+
+ assert second["cost_model"] == "gpt-4o"
+
+
def test_openai_compatible_adapter_uses_distinct_description_and_icon() -> None:
metadata = OpenAICompatibleLLMAdapter.get_metadata()