diff --git a/kubernetes/ai/llmkube/models/embed.yaml b/kubernetes/ai/llmkube/models/embed.yaml
index 9e65131a2..b68b9a148 100644
--- a/kubernetes/ai/llmkube/models/embed.yaml
+++ b/kubernetes/ai/llmkube/models/embed.yaml
@@ -22,7 +22,7 @@ spec:
   # CPU embedding service for the ToolHive gateway tool search.
   # Bound concurrent embedding work: eight slots and 2048-token batches
   # exceeded the 4Gi limit during ToolHive indexing.
-  image: ghcr.io/ggml-org/llama.cpp:server@sha256:af8c29600d1de84945de13fa7f126d06fdae16c4b19dc7db1c69c74287eae257
+  image: ghcr.io/ggml-org/llama.cpp:server@sha256:33868c035b21dc63f7c60b7438774283fd99215bc319114eb03de5df4ce7cd6b
   replicas: 1
   contextSize: 8192
   parallelSlots: 2