diff --git a/environment.yml b/environment.yml
index fffb4b14..6a8e6810 100644
--- a/environment.yml
+++ b/environment.yml
@@ -8,7 +8,7 @@ dependencies:
   - networkx
   - numpy
   - openff-toolkit-base >=0.15.1
-  - openff-units >=0.3.1
+  - openff-units >=0.3.2
   - openmm
   - pint
   - pip
@@ -22,3 +22,7 @@ dependencies:
   - typing-extensions
   - ipywidgets  # opt
   - zstandard
+  - pip:
+    - git+https://github.com/openforcefield/openff-units/@define-pydantic-schema
+    - git+https://github.com/openforcefield/openff-interchange/@fix-pydantic
+    - pydantic==2.13.*
diff --git a/src/gufe/settings/models.py b/src/gufe/settings/models.py
index 7627bc17..d7e85889 100644
--- a/src/gufe/settings/models.py
+++ b/src/gufe/settings/models.py
@@ -96,9 +96,11 @@ def __eq__(self, other: Any) -> bool:
         # which results in frozen/unfrozen objects not being equal
         # https://github.com/pydantic/pydantic/blob/2486e068e85c51728c9f2d344cfee2f7e11d555c/pydantic/v1/main.py#L911
         if isinstance(other, _BaseModel):
-            return self.model_dump() == other.model_dump()
+            return self.model_dump(polymorphic_serialization=True) == other.model_dump(polymorphic_serialization=True)
+            # TODO: as of pydantic 3.13, polymorphic_serialization is required, or else the default(?) will be used instead?
+            # https://pydantic.dev/docs/validation/latest/concepts/serialization#polymorphic-serialization
         else:
-            return self.model_dump() == other
+            return self.model_dump(polymorphic_serialization=True) == other
 
 
 class ThermoSettings(SettingsBaseModel):
diff --git a/src/gufe/tests/test_models.py b/src/gufe/tests/test_models.py
index ee613e03..302dd376 100644
--- a/src/gufe/tests/test_models.py
+++ b/src/gufe/tests/test_models.py
@@ -80,7 +80,7 @@ def test_settings_schema():
 
 
 def test_openmmffsettings_schema():
-    expected_schema = {
+    expected_val_schema = {
         "additionalProperties": False,
         "description": "Parameters to set up the force field with OpenMM ForceFields\n\n.. note::\n   Currently, this stores what is needed for the\n   :class:`openmmforcefields.system_generators.SystemGenerator` signature.\n   See the `OpenMMForceField SystemGenerator documentation`_ for more details.\n\n\n.. _`OpenMMForceField SystemGenerator documentation`:\n   https://github.com/openmm/openmmforcefields#automating-force-field-management-with-systemgenerator",
         "properties": {
@@ -110,6 +110,7 @@ def test_openmmffsettings_schema():
             },
             "nonbonded_method": {"default": "PME", "title": "Nonbonded Method", "type": "string"},
             "nonbonded_cutoff": {
+                "default": '{"magnitude": 0.9, "units": "nanometer"}',
                 "description": "Cutoff value for short range nonbonded interactions in nm. Compatible units will be converted to nm.",
                 "title": "Nonbonded Cutoff",
                 "type": "number",
@@ -118,10 +119,14 @@ def test_openmmffsettings_schema():
         "title": "OpenMMSystemGeneratorFFSettings",
         "type": "object",
     }
-    ser_schema = OpenMMSystemGeneratorFFSettings.model_json_schema(mode="serialization")
+
     val_schema = OpenMMSystemGeneratorFFSettings.model_json_schema(mode="validation")
-    assert ser_schema == expected_schema
-    assert val_schema == expected_schema
+    assert val_schema == expected_val_schema
+
+    ser_schema = OpenMMSystemGeneratorFFSettings.model_json_schema(mode="serialization")
+    expected_ser_schema = expected_val_schema.copy()
+    expected_ser_schema["properties"]["nonbonded_cutoff"]["default"] = {"val": 0.9, "unit": "nanometer"}
+    assert ser_schema == expected_ser_schema
 
 
 def test_default_settings():
diff --git a/src/gufe/vendor/openff/interchange/pydantic.py b/src/gufe/vendor/openff/interchange/pydantic.py
index 88853fd8..52d70f89 100644
--- a/src/gufe/vendor/openff/interchange/pydantic.py
+++ b/src/gufe/vendor/openff/interchange/pydantic.py
@@ -15,7 +15,7 @@ class _BaseModel(BaseModel):
     )
 
     def model_dump(self, **kwargs) -> dict[str, Any]:
-        return super().model_dump(serialize_as_any=True, **kwargs)
+        return super().model_dump(**kwargs)
 
     def model_dump_json(self, **kwargs) -> str:
-        return super().model_dump_json(serialize_as_any=True, **kwargs)
+        return super().model_dump_json(**kwargs)