diff --git a/ggml/src/ggml-webgpu/ggml-webgpu.cpp b/ggml/src/ggml-webgpu/ggml-webgpu.cpp
index 32e120266a..b29c0a8846 100644
--- a/ggml/src/ggml-webgpu/ggml-webgpu.cpp
+++ b/ggml/src/ggml-webgpu/ggml-webgpu.cpp
@@ -1168,7 +1168,8 @@ static webgpu_command ggml_webgpu_mul_mat(webgpu_context & ctx,
             uint32_t batches       = dst->ne[2] * dst->ne[3];
             uint32_t output_groups = CEIL_DIV(dst->ne[0], WEBGPU_MUL_MAT_VEC_OUTPUTS_PER_WG);
             uint32_t total_wg      = output_groups * batches;
-            wg_x                   = total_wg % ctx->global_ctx->capabilities.limits.maxComputeWorkgroupsPerDimension;
+            // TODO: split large sizes into multiple batches to avoid way over-provisioning workgroups
+            wg_x = std::min(total_wg, ctx->global_ctx->capabilities.limits.maxComputeWorkgroupsPerDimension);
             wg_y = CEIL_DIV(total_wg, ctx->global_ctx->capabilities.limits.maxComputeWorkgroupsPerDimension);
         } else {
             pipeline = ctx->mul_mat_pipelines[src0->type][src1->type][vectorized];