diff --git a/openpilot/selfdrive/modeld/models/driving_tinygrad.pkl.chunk01of02 b/openpilot/selfdrive/modeld/models/driving_tinygrad.pkl.chunk01of02
index 6782bba16..f5ff3d5d3 100644
Binary files a/openpilot/selfdrive/modeld/models/driving_tinygrad.pkl.chunk01of02 and b/openpilot/selfdrive/modeld/models/driving_tinygrad.pkl.chunk01of02 differ
diff --git a/openpilot/selfdrive/modeld/models/driving_tinygrad.pkl.chunk02of02 b/openpilot/selfdrive/modeld/models/driving_tinygrad.pkl.chunk02of02
index 2959a0c45..0b17418f5 100644
Binary files a/openpilot/selfdrive/modeld/models/driving_tinygrad.pkl.chunk02of02 and b/openpilot/selfdrive/modeld/models/driving_tinygrad.pkl.chunk02of02 differ
diff --git a/openpilot/sunnypilot/modeld_v2/modeld.py b/openpilot/sunnypilot/modeld_v2/modeld.py
index 6584ee4c9..e9469478b 100755
--- a/openpilot/sunnypilot/modeld_v2/modeld.py
+++ b/openpilot/sunnypilot/modeld_v2/modeld.py
@@ -75,10 +75,18 @@ def _pkl_exists(path):
   return all(os.path.exists(get_chunk_name(path, i, num_chunks)) for i in range(num_chunks))
 
 
-def _find_driving_pkl(bundle):
+def _find_driving_pkl(bundle, chestnut: bool = False):
   if (override := os.environ.get('COMBINED_MODEL_PKL')) and _pkl_exists(override):
     return override
-  if bundle is None or not bundle.models:
+  if bundle is None:
+    # no active bundle: use the bundled default pkl (CD210) so a fresh install
+    # works immediately without downloading a model
+    from openpilot.selfdrive.modeld.helpers import modeld_pkl_path
+    bundled = str(modeld_pkl_path(chestnut))
+    if _pkl_exists(bundled):
+      return bundled
+    return None
+  if not bundle.models:
     return None
   from openpilot.common.hardware.hw import Paths
   model_root = Paths.model_root()
@@ -121,7 +129,7 @@ class ModelState(ModelStateBase):
     self.PLANPLUS_CONTROL: float = 1.0
     self.chestnut = chestnut
 
-    pkl_path = _find_driving_pkl(model_bundle)
+    pkl_path = _find_driving_pkl(model_bundle, chestnut=chestnut)
     assert pkl_path is not None, f"No driving pkl found for {'chestnut' if chestnut else 'small model'} — all models must be compiled with compile_modeld.py"
     self._init_combined(pkl_path, cam_w, cam_h, model_bundle)
 
diff --git a/openpilot/sunnypilot/models/helpers.py b/openpilot/sunnypilot/models/helpers.py
index a1a392c5e..842595cec 100644
--- a/openpilot/sunnypilot/models/helpers.py
+++ b/openpilot/sunnypilot/models/helpers.py
@@ -175,6 +175,24 @@ def validate_active_bundles(params: Params, source_bundles: dict[str, list[custo
   get_active_model_runner(params, force_check=True)
 
 
+def _bundled_default_pkl_available() -> bool:
+  # the repo ships the default model (CD210) as driving_tinygrad.pkl chunks; only
+  # a real pkl or all chunk files count (a bare chunkmanifest does not)
+  from openpilot.common.file_chunker import get_chunk_name, get_manifest_path
+  from openpilot.selfdrive.modeld.helpers import modeld_pkl_path
+  path = str(modeld_pkl_path(chestnut=False))
+  if os.path.exists(path):
+    return True
+  manifest_path = get_manifest_path(path)
+  if not os.path.exists(manifest_path):
+    return False
+  try:
+    num_chunks = int(open(manifest_path).read().strip())
+  except Exception:
+    return False
+  return all(os.path.exists(get_chunk_name(path, i, num_chunks)) for i in range(num_chunks))
+
+
 def get_active_model_runner(params: Params | None = None, force_check: bool = False) -> int:
   params = params or Params()
   cached_runner_type = params.get("ModelRunnerTypeCache")
@@ -183,6 +201,10 @@ def get_active_model_runner(params: Params | None = None, force_check: bool = Fa
   runner_type = custom.ModelManagerSP.Runner.stock
   if active_bundle := get_active_bundle(params):
     runner_type = active_bundle.runner.raw
+  elif _bundled_default_pkl_available():
+    # no bundle selected: modeld_v2 runs the bundled default pkl (CD210), so it
+    # must not fall back to stock modeld (which expects a run_model-format pkl)
+    runner_type = custom.ModelManagerSP.Runner.tinygrad
 
   if cached_runner_type != runner_type:
     params.put("ModelRunnerTypeCache", int(runner_type), block=True)