diff --git a/changelog_entry.yaml b/changelog_entry.yaml
index e69de29b..179b568c 100644
--- a/changelog_entry.yaml
+++ b/changelog_entry.yaml
@@ -0,0 +1,5 @@
+- bump: patch
+  changes:
+    fixed:
+    - US model always downlaods from HuggingFace.
+    - Subsampling improvements.
diff --git a/policyengine_api/jobs/calculate_economy_simulation_job.py b/policyengine_api/jobs/calculate_economy_simulation_job.py
index c976db39..f2923a2c 100644
--- a/policyengine_api/jobs/calculate_economy_simulation_job.py
+++ b/policyengine_api/jobs/calculate_economy_simulation_job.py
@@ -20,7 +20,11 @@
 from policyengine_api.endpoints.economy.compare import compare_economic_outputs
 from policyengine_api.endpoints.economy.reform_impact import set_comment_on_job
 from policyengine_api.constants import COUNTRY_PACKAGE_VERSIONS
-from policyengine_api.country import COUNTRIES, create_policy_reform
+from policyengine_api.country import (
+    COUNTRIES,
+    create_policy_reform,
+    PolicyEngineCountry,
+)
 from policyengine_api.utils.v2_v1_comparison import (
     V2V1Comparison,
     compute_difference,
@@ -371,15 +375,13 @@ def _compute_economy(
                 options.get("max_households", os.environ.get("MAX_HOUSEHOLDS"))
                 is not None
             ):
-                simulation.subsample(
-                    int(
-                        options.get(
-                            "max_households",
-                            os.environ.get("MAX_HOUSEHOLDS", 1_000_000),
-                        )
-                    ),
-                    seed=(region, time_period),
+                simulation = subsample(
+                    options=options,
+                    simulation=simulation,
+                    region=region,
                     time_period=time_period,
+                    reform=reform,
+                    country=country,
                 )
             simulation.default_calculation_period = time_period
 
@@ -575,6 +577,44 @@ def _compute_cliff_impacts(self, simulation: Microsimulation) -> Dict:
         }
 
 
+def subsample(
+    options: dict,
+    simulation: Microsimulation,
+    region: str,
+    time_period: str,
+    reform: dict,
+    country: PolicyEngineCountry,
+) -> Microsimulation:
+    """
+    Subsamples a microsimulation dataset and reinitializes the simulation with the subsampled data.
+    Args:
+        options (dict): A dictionary of options, which may include "max_households" to specify the maximum number of households to subsample.
+        simulation (Microsimulation): The original microsimulation object to be subsampled.
+        region (str): The region for which the simulation is being run.
+        time_period (str): The time period for which the simulation is being run.
+        reform (dict): A dictionary representing the policy reform to apply to the simulation.
+        country (PolicyEngineCountry): The country-specific policy engine object.
+    Returns:
+        Microsimulation: A new microsimulation object initialized with the subsampled data and the specified reform.
+    """
+
+    simulation.subsample(
+        int(
+            options.get(
+                "max_households",
+                os.environ.get("MAX_HOUSEHOLDS", 1_000_000),
+            )
+        ),
+        seed=(region, time_period),
+        time_period=time_period,
+    )
+    input_data = simulation.to_input_dataframe()
+    simulation = country.country_package.Microsimulation(
+        dataset=input_data,
+        reform=reform,
+    )
+    return simulation
+
 class SimulationAPIv2:
     project: str
     location: str