apache-airflow-providers-google gates google-cloud-aiplatform[evaluation] to Python <3.14 (falling back to plain google-cloud-aiplatform on 3.14) because the evaluation extra pulls litellm, which cannot be installed on Python 3.14.
Root cause: google-cloud-aiplatform (latest 1.163.0) caps its evaluation extra at litellm >=1.83.7,<1.86.0. No litellm in that range supports 3.14 (the only 3.14-installable one, 1.83.7, pins pydantic==2.12.5 which is incompatible with 3.14). litellm re-added 3.14 support at 1.95.0, but that is above aiplatform's <1.86.0 ceiling. So pip install apache-airflow-providers-google fails with ResolutionImpossible on 3.14, despite the provider advertising requires_python >=3.10.
Follow-up (to remove the workaround): once google-cloud-aiplatform raises its evaluation-extra litellm ceiling to a 3.14-capable version (>=1.95), remove the python_version split in providers/google/pyproject.toml and restore the single google-cloud-aiplatform[evaluation] line.
apache-airflow-providers-google gates google-cloud-aiplatform[evaluation] to Python <3.14 (falling back to plain google-cloud-aiplatform on 3.14) because the evaluation extra pulls litellm, which cannot be installed on Python 3.14.
Root cause: google-cloud-aiplatform (latest 1.163.0) caps its evaluation extra at litellm >=1.83.7,<1.86.0. No litellm in that range supports 3.14 (the only 3.14-installable one, 1.83.7, pins pydantic==2.12.5 which is incompatible with 3.14). litellm re-added 3.14 support at 1.95.0, but that is above aiplatform's <1.86.0 ceiling. So pip install apache-airflow-providers-google fails with ResolutionImpossible on 3.14, despite the provider advertising requires_python >=3.10.
Follow-up (to remove the workaround): once google-cloud-aiplatform raises its evaluation-extra litellm ceiling to a 3.14-capable version (>=1.95), remove the python_version split in providers/google/pyproject.toml and restore the single google-cloud-aiplatform[evaluation] line.