Nightly OSS: usage.prompt_tokens_details=None

NIGHTLY_CODEX_FINAL_ATTEMPT_1.md

@@ -0,0 +1,15 @@
+Implemented a small related contribution for openai/openai-python issue #2544.
+
+Changed:
+- [tests/test_models.py](/home/runner/work/oss-nightly-control/oss-nightly-control/target/tests/test_models.py): added a regression test that `CompletionUsage` preserves provider-specific extra usage fields like `promptTokensDetails`.
+- [README.md](/home/runner/work/oss-nightly-control/oss-nightly-control/target/README.md): clarified nested undocumented response fields are available from the nested Pydantic model.
+- [NIGHTLY_REPORT.md](/home/runner/work/oss-nightly-control/oss-nightly-control/target/NIGHTLY_REPORT.md): added the required nightly summary.
+
+Validation run:
+- `uv run pytest tests/test_models.py::test_completion_usage_preserves_unknown_token_details -q -o addopts=''`
+- `uv run pytest tests/test_models.py -q -o addopts=''`
+- `uv run ruff format --check tests/test_models.py`
+- `uv run ruff check tests/test_models.py`
+- `uv run python scripts/utils/ruffen-docs.py README.md`
+
+Note: I avoided the direct `prompt_tokens_details=None` behavior change because GitHub shows an existing open PR for that exact approach: https://github.com/openai/openai-python/pull/2773. The issue context used was https://github.com/openai/openai-python/issues/2544.

README.md

@@ -848,6 +848,7 @@ options.
 To access undocumented response properties, you can access the extra fields like `response.unknown_prop`. You
 can also get all the extra fields on the Pydantic model as a dict with
 [`response.model_extra`](https://docs.pydantic.dev/latest/api/base_model/#pydantic.BaseModel.model_extra).
+For nested response objects, access extra fields from that nested model, such as `response.usage.model_extra`.
 
 ### Configuring the HTTP client
 

tests/test_models.py

@@ -7,6 +7,7 @@
 import pydantic
 from pydantic import Field
 
+from openai.types import CompletionUsage
 from openai._utils import PropertyInfo
 from openai._compat import PYDANTIC_V1, parse_obj, model_dump, model_json
 from openai._models import DISCRIMINATOR_CACHE, BaseModel, construct_type
@@ -157,6 +158,36 @@ def test_unknown_fields() -> None:
     assert model_dump(m2) == {"foo": "foo", "unknown": {"foo_bar": True}}
 
 
+def test_completion_usage_preserves_unknown_token_details() -> None:
+    usage = parse_obj(
+        CompletionUsage,
+        {
+            "completion_tokens": 57,
+            "prompt_tokens": 2181,
+            "total_tokens": 2518,
+            "prompt_tokens_details": None,
+            "reasoning_tokens": 280,
+            "traffic_type": "ON_DEMAND",
+            "promptTokensDetails": [{"modality": "TEXT", "tokenCount": 2181}],
+            "candidatesTokensDetails": [{"modality": "TEXT", "tokenCount": 57}],
+        },
+    )
+
+    assert usage.prompt_tokens_details is None
+    assert cast(Any, usage).reasoning_tokens == 280
+    assert cast(Any, usage).traffic_type == "ON_DEMAND"
+    assert cast(Any, usage).promptTokensDetails == [{"modality": "TEXT", "tokenCount": 2181}]
+    assert usage.to_dict(exclude_none=True) == {
+        "completion_tokens": 57,
+        "prompt_tokens": 2181,
+        "total_tokens": 2518,
+        "reasoning_tokens": 280,
+        "traffic_type": "ON_DEMAND",
+        "promptTokensDetails": [{"modality": "TEXT", "tokenCount": 2181}],
+        "candidatesTokensDetails": [{"modality": "TEXT", "tokenCount": 57}],
+    }
+
+
 def test_strict_validation_unknown_fields() -> None:
     class Model(BaseModel):
         foo: str