diff --git a/vllm/model_executor/kernels/linear/scaled_mm/flashinfer.py b/vllm/model_executor/kernels/linear/scaled_mm/flashinfer.py
index 72a3b84..8debafa 100644
--- a/vllm/model_executor/kernels/linear/scaled_mm/flashinfer.py
+++ b/vllm/model_executor/kernels/linear/scaled_mm/flashinfer.py
@@ -153,8 +153,8 @@ class FlashInferFp8DeepGEMMDynamicBlockScaledKernel(
     Conditional FlashInfer / DeepGEMM FP8 block-scaled GEMM.
 
     Dispatches between two kernels based on input batch size:
-    - Small batches (M < 32): FlashInfer's swapAB trick for better utilisation.
-    - Large batches (M >= 32): DeepGEMM for peak throughput.
+    - Decode-sized batches (M <= 32): FlashInfer's fused BF16-to-FP8 path.
+    - Larger batches (M > 32): DeepGEMM for peak throughput.
 
     apply_input_quant is False because FlashInfer accepts BF16 input and
     handles FP8 conversion internally.  The DeepGEMM branch therefore
@@ -240,8 +240,8 @@ def _dynamic_flashinfer_deepgemm_blockscale_gemm_impl(
     Conditional FlashInfer FP8 blockscale GEMM with batch-size-dependent selection.
 
     This function switches between two optimized kernels based on the input batch size:
-    - For small batches (M < 32): Uses FlashInfer's DeepGEMM swapAB optimization.
-    - For larger batches (M >= 32): Uses the official DeepGEMM kernel.
+    - For decode-sized batches (M <= 32): Uses FlashInfer's fused FP8 path.
+    - For larger batches (M > 32): Uses the official DeepGEMM kernel.
 
     The conditional logic must use torch.cond() instead of a simple if-else statement
     to maintain compatibility with torch.compile graph compilation.
@@ -301,12 +301,12 @@ def _dynamic_flashinfer_deepgemm_blockscale_gemm_impl(
     if envs.VLLM_BATCH_INVARIANT:
         return run_deepgemm(input, weight, weight_scale)
 
-    condition = input.shape[0] < 32
+    condition = input.shape[0] <= 32
 
     # PyTorch's torch.compile cannot handle input-dependent control flow in standard
     # Python conditionals. torch.cond() explicitly registers both code paths in the
     # computation graph, allowing torch.compile to capture both branches.
-    # without torch.cond, the M < 32 condition won't be able to be captured by torch
+    # without torch.cond, the M <= 32 condition won't be able to be captured by torch
     # compile
     return torch.cond(
         condition,