diff --git a/test/external/external_model_benchmark.py b/test/external/external_model_benchmark.py
index 2edf18c9f8..52ebd15e23 100644
--- a/test/external/external_model_benchmark.py
+++ b/test/external/external_model_benchmark.py
@@ -12,10 +12,10 @@ from tinygrad.tensor import Tensor
 from tinygrad import Device
 
 MODELS = {
-  "resnet50": "https://github.com/onnx/models/raw/main/archive/vision/classification/resnet/model/resnet50-caffe2-v1-9.onnx",
+  "resnet50": "https://github.com/onnx/models/raw/main/validated/vision/classification/resnet/model/resnet50-caffe2-v1-9.onnx",
   "openpilot": "https://github.com/commaai/openpilot/raw/v0.9.4/selfdrive/modeld/models/supercombo.onnx",
-  "efficientnet": "https://github.com/onnx/models/raw/main/archive/vision/classification/efficientnet-lite4/model/efficientnet-lite4-11.onnx",
-  "shufflenet": "https://github.com/onnx/models/raw/main/archive/vision/classification/shufflenet/model/shufflenet-9.onnx",
+  "efficientnet": "https://github.com/onnx/models/raw/main/validated/vision/classification/efficientnet-lite4/model/efficientnet-lite4-11.onnx",
+  "shufflenet": "https://github.com/onnx/models/raw/main/validated/vision/classification/shufflenet/model/shufflenet-9.onnx",
   "commavq": "https://huggingface.co/commaai/commavq-gpt2m/resolve/main/gpt2m.onnx",
 
   # broken in torch MPS
diff --git a/test/models/test_onnx.py b/test/models/test_onnx.py
index afa9e3c9aa..db7b708823 100644
--- a/test/models/test_onnx.py
+++ b/test/models/test_onnx.py
@@ -99,17 +99,17 @@ class TestOnnxModel(unittest.TestCase):
   @unittest.skip("slow")
   def test_efficientnet(self):
     input_name, input_new = "images:0", True
-    self._test_model(fetch("https://github.com/onnx/models/raw/main/archive/vision/classification/efficientnet-lite4/model/efficientnet-lite4-11.onnx"), input_name, input_new)   # noqa: E501
+    self._test_model(fetch("https://github.com/onnx/models/raw/main/validated/vision/classification/efficientnet-lite4/model/efficientnet-lite4-11.onnx"), input_name, input_new)   # noqa: E501
 
   def test_shufflenet(self):
     input_name, input_new = "gpu_0/data_0", False
-    self._test_model(fetch("https://github.com/onnx/models/raw/main/archive/vision/classification/shufflenet/model/shufflenet-9.onnx"), input_name, input_new)   # noqa: E501
+    self._test_model(fetch("https://github.com/onnx/models/raw/main/validated/vision/classification/shufflenet/model/shufflenet-9.onnx"), input_name, input_new)   # noqa: E501
 
   @unittest.skip("test is very slow")
   def test_resnet(self):
     # NOTE: many onnx models can't be run right now due to max pool with strides != kernel_size
     input_name, input_new = "data", False
-    self._test_model(fetch("https://github.com/onnx/models/raw/main/archive/vision/classification/resnet/model/resnet18-v2-7.onnx"), input_name, input_new)   # noqa: E501
+    self._test_model(fetch("https://github.com/onnx/models/raw/main/validated/vision/classification/resnet/model/resnet18-v2-7.onnx"), input_name, input_new)   # noqa: E501
 
   def _test_model(self, fn, input_name, input_new, debug=False):
     onnx_model = onnx.load(fn)