diff --git a/.ci/scripts/gather_test_models.py b/.ci/scripts/gather_test_models.py
index 80f451f5f4f..32d8535ac7e 100755
--- a/.ci/scripts/gather_test_models.py
+++ b/.ci/scripts/gather_test_models.py
@@ -31,6 +31,7 @@
         "llama3_2_text_decoder": "linux.4xlarge.memory",
         # This one causes timeout on smaller runner, the root cause is unclear (T161064121)
         "dl3": "linux.4xlarge.memory",
+        "edsr": "linux.4xlarge.memory",
         "emformer_join": "linux.4xlarge.memory",
         "emformer_predict": "linux.4xlarge.memory",
         "phi_4_mini": "linux.4xlarge.memory",
@@ -102,6 +103,19 @@ def model_should_run_on_event(model: str, event: str) -> bool:
         return True
 
 
+def model_should_run_on_backend(model: str, backend: str) -> bool:
+    """
+    A helper function to decide whether a model should be tested on a backend.
+    """
+    # Exporting dl3 with the portable backend does not finish inside the 90 minute
+    # job limit, not even on the bigger runner it already uses, so the job only
+    # ever reports cancelled. Its xnnpack variant still covers the model.
+    # TODO(#21692): re-enable once portable export of dl3 fits in the limit.
+    if backend == "portable" and model == "dl3":
+        return False
+    return True
+
+
 def model_should_run_on_target_os(model: str, target_os: str) -> bool:
     """
     A helper function to decide whether a model should be tested on a target os (linux/macos).
@@ -153,7 +167,9 @@ def export_models_for_ci() -> dict[str, dict]:
         if not model_should_run_on_event(name, event):
             continue
 
-        if not model_should_run_on_target_os(name, target_os):
+        if not model_should_run_on_target_os(
+            name, target_os
+        ) or not model_should_run_on_backend(name, backend):
             continue
 
         if backend == "xnnpack":