diff --git a/.github/workflows/java.yml b/.github/workflows/java.yml
index b0eacd30fbe..a4a5d6c5bd6 100644
--- a/.github/workflows/java.yml
+++ b/.github/workflows/java.yml
@@ -65,13 +65,18 @@ jobs:
           - spark: 4.1.1
             scala: 2.13.17
             jdk: '17'
+            skipLibPostalTests: 'true'
           - spark: 4.0.2
             scala: 2.13.17
             jdk: '17'
+            skipLibPostalTests: 'true'
           - spark: 3.5.8
             scala: 2.13.8
             jdk: '11'
             skipTests: ''
+            skipLibPostalTests: 'true'
+          # The libpostal suite needs ~1.3 GB of model data, so it runs in this
+          # combination only. See the libpostal steps below.
           - spark: 3.5.8
             scala: 2.12.15
             jdk: '11'
@@ -80,10 +85,12 @@ jobs:
             scala: 2.13.8
             jdk: '11'
             skipTests: ''
+            skipLibPostalTests: 'true'
           - spark: 3.4.0
             scala: 2.12.15
             jdk: '11'
             skipTests: ''
+            skipLibPostalTests: 'true'
     steps:
       - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
         with:
@@ -101,10 +108,45 @@ jobs:
           path: ~/.m2
           key: ${{ runner.os }}-m2-${{ hashFiles('**/pom.xml') }}
           restore-keys: ${{ runner.os }}-m2
+      # jpostal downloads ~1.3 GB of model data the first time an address
+      # function runs. Cache it so the suite does not depend on the object
+      # store being reachable. The key is a constant because the data version
+      # is pinned by the jpostal dependency, not by anything in the tree; bump
+      # it when that dependency changes.
+      - name: Cache libpostal model data
+        if: ${{ !matrix.skipLibPostalTests }}
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: /tmp/libpostal
+          key: libpostal-senzing-v1.1.0-jpostal-1.2.2
+      # On a cache miss, fetch the data here rather than letting jpostal do it
+      # inside the test JVM: DataDownloadUtils has a 15 s connect timeout and
+      # no retry, so a transient stall fails the whole build.
+      - name: Pre-fetch libpostal model data
+        if: ${{ !matrix.skipLibPostalTests }}
+        run: |
+          set -euo pipefail
+          DIR=/tmp/libpostal
+          populated=1
+          for d in transliteration numex address_parser address_expansions language_classifier; do
+            [ -d "$DIR/$d" ] || populated=0
+          done
+          if [ "$populated" = 1 ]; then
+            echo "libpostal data already present, skipping download"
+            exit 0
+          fi
+          mkdir -p "$DIR"
+          for f in libpostal_data.tar.gz parser.tar.gz language_classifier.tar.gz; do
+            curl -fsSL --retry 5 --retry-all-errors --retry-delay 10 --connect-timeout 30 \
+              -o "$DIR/$f" "https://public-read-libpostal-data.s3.amazonaws.com/v1.1.0/$f"
+            tar -xzf "$DIR/$f" -C "$DIR"
+            rm -f "$DIR/$f"
+          done
       - env:
           SPARK_VERSION: ${{ matrix.spark }}
           SCALA_VERSION: ${{ matrix.scala }}
           SKIP_TESTS: ${{ matrix.skipTests }}
+          SEDONA_SKIP_LIBPOSTAL_TESTS: ${{ matrix.skipLibPostalTests }}
         run: |
           SPARK_COMPAT_VERSION=${SPARK_VERSION:0:3}
 
diff --git a/spark/common/src/test/scala/org/apache/sedona/sql/AddressProcessingFunctionsTest.scala b/spark/common/src/test/scala/org/apache/sedona/sql/AddressProcessingFunctionsTest.scala
index 6b13d4b909f..45f2cb9fa52 100644
--- a/spark/common/src/test/scala/org/apache/sedona/sql/AddressProcessingFunctionsTest.scala
+++ b/spark/common/src/test/scala/org/apache/sedona/sql/AddressProcessingFunctionsTest.scala
@@ -18,40 +18,26 @@
  */
 package org.apache.sedona.sql
 
-import org.apache.sedona.core.utils.SedonaConf
 import org.apache.spark.sql.sedona_sql.expressions.st_functions.{ExpandAddress, ParseAddress}
 import org.apache.spark.sql.{Row, functions => f}
-import org.scalatest.BeforeAndAfterEach
-import org.slf4j.LoggerFactory
+import org.scalatest.{Canceled, Outcome}
 
-import java.nio.file.{Files, Paths}
 import scala.collection.mutable
 
-class AddressProcessingFunctionsTest extends TestBaseScala with BeforeAndAfterEach {
-  private val logger = LoggerFactory.getLogger(getClass)
-  var clearedLibPostal = false
-
-  def clearLibpostalDataDir(): String = {
-    val dir = SedonaConf.fromActiveSession().getLibPostalDataDir
-    if (dir != null && dir.nonEmpty) {
-      try {
-        Files
-          .walk(Paths.get(dir))
-          .sorted(java.util.Comparator.reverseOrder())
-          .forEach(path => Files.deleteIfExists(path))
-      } catch {
-        case e: Exception =>
-          logger.warn(s"Failed to clear libpostal data directory: ${e.getMessage}")
-      }
-    }
-    dir
-  }
+class AddressProcessingFunctionsTest extends TestBaseScala {
+
+  // These tests need ~1.3 GB of libpostal model data, which jpostal downloads
+  // on first use. CI runs them in a single matrix combination and sets this
+  // variable everywhere else; unset, the tests run as usual.
+  private val skipLibPostalTests =
+    sys.env.get("SEDONA_SKIP_LIBPOSTAL_TESTS").exists(_.equalsIgnoreCase("true"))
 
-  override def beforeEach(): Unit = {
-    super.beforeEach()
-    if (!clearedLibPostal) {
-      clearedLibPostal = true
-      clearLibpostalDataDir()
+  override def withFixture(test: NoArgTest): Outcome = {
+    if (skipLibPostalTests) {
+      Canceled(
+        "Skipping libpostal test - unset SEDONA_SKIP_LIBPOSTAL_TESTS to download the model data and run")
+    } else {
+      super.withFixture(test)
     }
   }