diff --git a/recipes/nemotron-3-nano-nvfp4.yaml b/recipes/nemotron-3-nano-nvfp4.yaml
index c835e45..43f1383 100644
--- a/recipes/nemotron-3-nano-nvfp4.yaml
+++ b/recipes/nemotron-3-nano-nvfp4.yaml
@@ -27,15 +27,10 @@ defaults:
   gpu_memory_utilization: 0.7
   max_model_len: 262144
 
-# Environment variables
-env:
-  VLLM_NVFP4_GEMM_BACKEND: "marlin"
-  VLLM_TEST_FORCE_FP8_MARLIN: "1"
-  VLLM_MARLIN_USE_ATOMIC_ADD: "1"
-
 # The vLLM serve command template
 command: |
   vllm serve nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4  \
+     --moe-backend cutlass \
      --max-model-len {max_model_len} \
      --port {port} --host {host} \
      --trust-remote-code \
diff --git a/recipes/nemotron-3-super-nvfp4.yaml b/recipes/nemotron-3-super-nvfp4.yaml
index a4de32d..ec790c2 100644
--- a/recipes/nemotron-3-super-nvfp4.yaml
+++ b/recipes/nemotron-3-super-nvfp4.yaml
@@ -1,8 +1,8 @@
 # Recipe: Nemotron-3-Super-NVFP4
-# Optimized for Marlin backend throughput
+# Uses VLLM_CUTLASS for NVFP4
 recipe_version: "1"
-name: Nemotron-3-Super-NVFP4-Marlin-Optimized
-description: vLLM serving Nemotron-3-Super-120B using Marlin kernels
+name: Nemotron-3-Super-NVFP4-CUTLASS-Optimized
+description: vLLM serving Nemotron-3-Super-120B using CUTLASS kernels
 
 model: nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4
 container: vllm-node
@@ -20,15 +20,11 @@ defaults:
   gpu_memory_utilization: 0.7
   max_model_len: 262144
   max_num_seqs: 10
-env:
-  VLLM_NVFP4_GEMM_BACKEND: "marlin"
-  VLLM_TEST_FORCE_FP8_MARLIN: "1"
-  VLLM_MARLIN_USE_ATOMIC_ADD: "1"
 
 command: |
   vllm serve nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 \
   --kv-cache-dtype fp8 \
-  -tp {tensor_parallel} \
+  --moe-backend cutlass \
   --trust-remote-code \
   --gpu-memory-utilization {gpu_memory_utilization} \
   --max-model-len {max_model_len} \