diff --git a/hypencoder/requirements.txt b/hypencoder/requirements.txt
index 0795d5068..70a51fd54 100644
--- a/hypencoder/requirements.txt
+++ b/hypencoder/requirements.txt
@@ -6,7 +6,7 @@
 # Core ML stack. transformers must stay <5 - the Hypencoder checkpoints'
 # saved config triggers a meta-device path in transformers >=5 that
 # .from_pretrained() rejects. 4.45+ is fine.
-torch>=2.12.1
+torch>=2.13.0
 transformers>=5.0.0,<6 # auto updated from transformers>=4.45,<5 to avoid pointing at vulnerable versions
 
 # ONNX export + runtime (used both for export and for the optional INT8
diff --git a/reverse-image-search/script/requirements.txt b/reverse-image-search/script/requirements.txt
index c92167e47..ee90c3f3f 100644
--- a/reverse-image-search/script/requirements.txt
+++ b/reverse-image-search/script/requirements.txt
@@ -1,4 +1,4 @@
-torch>=2.12.1
+torch>=2.13.0
 transformers>=5.0.0
 torchvision>=0.20.0
 pillow>=11.0.0
diff --git a/visual-retrieval-colpali/src/legacy-requirements.txt b/visual-retrieval-colpali/src/legacy-requirements.txt
index 818e834de..123de712e 100644
--- a/visual-retrieval-colpali/src/legacy-requirements.txt
+++ b/visual-retrieval-colpali/src/legacy-requirements.txt
@@ -467,7 +467,7 @@ threadpoolctl==3.6.0
     # via scikit-learn
 tokenizers==0.23.1
     # via transformers
-torch==2.12.1
+torch==2.13.0
     # via
     #   visual-retrieval-colpali (pyproject.toml)
     #   accelerate