diff --git a/experimental/webgpu/webgpu_device.c b/experimental/webgpu/webgpu_device.c
index 4d26911..c9a2457 100644
--- a/experimental/webgpu/webgpu_device.c
+++ b/experimental/webgpu/webgpu_device.c
@@ -470,6 +470,7 @@
     .queue_alloca = iree_hal_webgpu_device_queue_alloca,
     .queue_dealloca = iree_hal_webgpu_device_queue_dealloca,
     .queue_fill = iree_hal_device_queue_emulated_fill,
+    .queue_update = iree_hal_device_queue_emulated_update,
     .queue_copy = iree_hal_device_queue_emulated_copy,
     .queue_read = iree_hal_webgpu_device_queue_read,
     .queue_write = iree_hal_webgpu_device_queue_write,
diff --git a/runtime/src/iree/hal/command_buffer.c b/runtime/src/iree/hal/command_buffer.c
index 58fda97..7f26263 100644
--- a/runtime/src/iree/hal/command_buffer.c
+++ b/runtime/src/iree/hal/command_buffer.c
@@ -662,6 +662,15 @@
               transfer_command->fill.pattern,
               transfer_command->fill.pattern_length, IREE_HAL_FILL_FLAG_NONE);
           break;
+        case IREE_HAL_TRANSFER_COMMAND_TYPE_UPDATE:
+          status = iree_hal_command_buffer_update_buffer(
+              command_buffer, transfer_command->update.source_buffer,
+              transfer_command->update.source_offset,
+              iree_hal_make_buffer_ref(transfer_command->update.target_buffer,
+                                       transfer_command->update.target_offset,
+                                       transfer_command->update.length),
+              IREE_HAL_UPDATE_FLAG_NONE);
+          break;
         case IREE_HAL_TRANSFER_COMMAND_TYPE_COPY:
           status = iree_hal_command_buffer_copy_buffer(
               command_buffer,
@@ -673,15 +682,6 @@
                                        transfer_command->copy.length),
               IREE_HAL_COPY_FLAG_NONE);
           break;
-        case IREE_HAL_TRANSFER_COMMAND_TYPE_UPDATE:
-          status = iree_hal_command_buffer_update_buffer(
-              command_buffer, transfer_command->update.source_buffer,
-              transfer_command->update.source_offset,
-              iree_hal_make_buffer_ref(transfer_command->update.target_buffer,
-                                       transfer_command->update.target_offset,
-                                       transfer_command->update.length),
-              IREE_HAL_UPDATE_FLAG_NONE);
-          break;
         default:
           status =
               iree_make_status(IREE_STATUS_INVALID_ARGUMENT,
diff --git a/runtime/src/iree/hal/command_buffer.h b/runtime/src/iree/hal/command_buffer.h
index 9152106..f15f74a 100644
--- a/runtime/src/iree/hal/command_buffer.h
+++ b/runtime/src/iree/hal/command_buffer.h
@@ -230,7 +230,7 @@
   IREE_HAL_FILL_FLAG_NONE = 0,
 };
 
-// Bitfield specifying flags controlling a update operation.
+// Bitfield specifying flags controlling an update operation.
 typedef uint64_t iree_hal_update_flags_t;
 enum iree_hal_update_flag_bits_t {
   IREE_HAL_UPDATE_FLAG_NONE = 0,
@@ -802,10 +802,10 @@
 typedef enum iree_hal_transfer_command_type_t {
   // iree_hal_command_buffer_fill_buffer
   IREE_HAL_TRANSFER_COMMAND_TYPE_FILL = 0u,
-  // iree_hal_command_buffer_copy_buffer
-  IREE_HAL_TRANSFER_COMMAND_TYPE_COPY = 1u,
   // iree_hal_command_buffer_update_buffer
-  IREE_HAL_TRANSFER_COMMAND_TYPE_UPDATE = 2u,
+  IREE_HAL_TRANSFER_COMMAND_TYPE_UPDATE = 1u,
+  // iree_hal_command_buffer_copy_buffer
+  IREE_HAL_TRANSFER_COMMAND_TYPE_COPY = 2u,
 } iree_hal_transfer_command_type_t;
 
 // Represents a single transfer command within a batch of commands.
@@ -821,14 +821,6 @@
       const void* pattern;
       iree_host_size_t pattern_length;
     } fill;
-    // IREE_HAL_TRANSFER_COMMAND_TYPE_COPY
-    struct {
-      iree_hal_buffer_t* source_buffer;
-      iree_device_size_t source_offset;
-      iree_hal_buffer_t* target_buffer;
-      iree_device_size_t target_offset;
-      iree_device_size_t length;
-    } copy;
     // IREE_HAL_TRANSFER_COMMAND_TYPE_UPDATE
     struct {
       const void* source_buffer;
@@ -837,6 +829,14 @@
       iree_device_size_t target_offset;
       iree_device_size_t length;
     } update;
+    // IREE_HAL_TRANSFER_COMMAND_TYPE_COPY
+    struct {
+      iree_hal_buffer_t* source_buffer;
+      iree_device_size_t source_offset;
+      iree_hal_buffer_t* target_buffer;
+      iree_device_size_t target_offset;
+      iree_device_size_t length;
+    } copy;
   };
 } iree_hal_transfer_command_t;
 
diff --git a/runtime/src/iree/hal/device.c b/runtime/src/iree/hal/device.c
index 5923ba4..f2b7d78 100644
--- a/runtime/src/iree/hal/device.c
+++ b/runtime/src/iree/hal/device.c
@@ -196,6 +196,91 @@
   return status;
 }
 
+IREE_API_EXPORT iree_status_t iree_hal_device_queue_emulated_update(
+    iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
+    const iree_hal_semaphore_list_t wait_semaphore_list,
+    const iree_hal_semaphore_list_t signal_semaphore_list,
+    const void* source_buffer, iree_host_size_t source_offset,
+    iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset,
+    iree_device_size_t length, iree_hal_update_flags_t flags) {
+  IREE_ASSERT_ARGUMENT(device);
+  IREE_ASSERT_ARGUMENT(source_buffer);
+  IREE_ASSERT_ARGUMENT(target_buffer);
+  IREE_TRACE_ZONE_BEGIN(z0);
+  IREE_TRACE_ZONE_APPEND_VALUE_I64(z0, (int64_t)length);
+
+  // If we are starting execution immediately then we can reduce latency by
+  // allowing inline command buffer execution.
+  iree_hal_command_buffer_mode_t command_buffer_mode =
+      IREE_HAL_COMMAND_BUFFER_MODE_ONE_SHOT;
+  if (wait_semaphore_list.count == 0) {
+    command_buffer_mode |= IREE_HAL_COMMAND_BUFFER_MODE_ALLOW_INLINE_EXECUTION;
+  }
+
+  // TODO(benvanik): support splitting the update into multiple chunks to fit
+  // under the max command buffer update size limit. This provisional API is
+  // intended only for updating dispatch parameters today.
+  if (length > UINT16_MAX) {
+    return iree_make_status(IREE_STATUS_UNIMPLEMENTED,
+                            "queue buffer updates currently limited to 64KB, "
+                            "tried to update %" PRIhsz " bytes",
+                            length);
+  }
+
+  iree_hal_transfer_command_t command = {
+      .type = IREE_HAL_TRANSFER_COMMAND_TYPE_UPDATE,
+      .update =
+          {
+              .source_buffer = source_buffer,
+              .source_offset = source_offset,
+              .target_buffer = target_buffer,
+              .target_offset = target_offset,
+              .length = length,
+          },
+  };
+
+  iree_hal_command_buffer_t* command_buffer = NULL;
+  IREE_RETURN_AND_END_ZONE_IF_ERROR(
+      z0, iree_hal_create_transfer_command_buffer(device, command_buffer_mode,
+                                                  queue_affinity, 1, &command,
+                                                  &command_buffer));
+
+  iree_status_t status = iree_hal_device_queue_execute(
+      device, queue_affinity, wait_semaphore_list, signal_semaphore_list,
+      command_buffer, iree_hal_buffer_binding_table_empty());
+
+  iree_hal_command_buffer_release(command_buffer);
+
+  IREE_TRACE_ZONE_END(z0);
+  return status;
+}
+
+IREE_API_EXPORT iree_status_t iree_hal_device_queue_update(
+    iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
+    const iree_hal_semaphore_list_t wait_semaphore_list,
+    const iree_hal_semaphore_list_t signal_semaphore_list,
+    const void* source_buffer, iree_host_size_t source_offset,
+    iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset,
+    iree_device_size_t length, iree_hal_update_flags_t flags) {
+  IREE_ASSERT_ARGUMENT(device);
+  IREE_ASSERT_ARGUMENT(
+      !wait_semaphore_list.count ||
+      (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values));
+  IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count ||
+                       (signal_semaphore_list.semaphores &&
+                        signal_semaphore_list.payload_values));
+  IREE_ASSERT_ARGUMENT(source_buffer);
+  IREE_ASSERT_ARGUMENT(target_buffer);
+  IREE_TRACE_ZONE_BEGIN(z0);
+  IREE_TRACE_ZONE_APPEND_VALUE_I64(z0, (int64_t)length);
+  iree_status_t status = _VTABLE_DISPATCH(device, queue_update)(
+      device, queue_affinity, wait_semaphore_list, signal_semaphore_list,
+      source_buffer, source_offset, target_buffer, target_offset, length,
+      flags);
+  IREE_TRACE_ZONE_END(z0);
+  return status;
+}
+
 IREE_API_EXPORT iree_status_t iree_hal_device_queue_emulated_copy(
     iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
     const iree_hal_semaphore_list_t wait_semaphore_list,
diff --git a/runtime/src/iree/hal/device.h b/runtime/src/iree/hal/device.h
index 145537a..4511fa7 100644
--- a/runtime/src/iree/hal/device.h
+++ b/runtime/src/iree/hal/device.h
@@ -312,11 +312,12 @@
     iree_hal_buffer_t* buffer);
 
 // Enqueues a single queue-ordered fill operation.
+// The |target_buffer| must be visible to the device queue performing the fill.
 //
 // WARNING: individual fills have a high overhead and batching should be
 // performed by the caller instead of calling this multiple times. The
 // iree_hal_create_transfer_command_buffer utility makes it easy to create
-// batches of transfer operations (fill, copy, update) and is only a few lines
+// batches of transfer operations (fill, update, copy) and is only a few lines
 // more code.
 IREE_API_EXPORT iree_status_t iree_hal_device_queue_fill(
     iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
@@ -326,12 +327,36 @@
     iree_device_size_t length, const void* pattern,
     iree_host_size_t pattern_length, iree_hal_fill_flags_t flags);
 
-// Enqueues a single queue-ordered copy operation.
+// Enqueues a single queue-ordered buffer update operation.
+// The provided |source_buffer| will be captured and need not remain live or
+// unchanged while the operation is queued. The |target_buffer| must be visible
+// to the device queue performing the update.
+//
+// Some implementations may have limits on the size of the update or may perform
+// poorly if the size is larger than an implementation-defined limit. Updates
+// should be kept as small and infrequent as possible.
 //
 // WARNING: individual copies have a high overhead and batching should be
 // performed by the caller instead of calling this multiple times. The
 // iree_hal_create_transfer_command_buffer utility makes it easy to create
-// batches of transfer operations (fill, copy, update) and is only a few lines
+// batches of transfer operations (fill, update, copy) and is only a few lines
+// more code.
+IREE_API_EXPORT iree_status_t iree_hal_device_queue_update(
+    iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
+    const iree_hal_semaphore_list_t wait_semaphore_list,
+    const iree_hal_semaphore_list_t signal_semaphore_list,
+    const void* source_buffer, iree_host_size_t source_offset,
+    iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset,
+    iree_device_size_t length, iree_hal_update_flags_t flags);
+
+// Enqueues a single queue-ordered copy operation.
+// The |source_buffer| and |target_buffer| must both be visible to the device
+// queue performing the copy.
+//
+// WARNING: individual copies have a high overhead and batching should be
+// performed by the caller instead of calling this multiple times. The
+// iree_hal_create_transfer_command_buffer utility makes it easy to create
+// batches of transfer operations (fill, update, copy) and is only a few lines
 // more code.
 IREE_API_EXPORT iree_status_t iree_hal_device_queue_copy(
     iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
@@ -578,6 +603,14 @@
       iree_device_size_t length, const void* pattern,
       iree_host_size_t pattern_length, iree_hal_fill_flags_t flags);
 
+  iree_status_t(IREE_API_PTR* queue_update)(
+      iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
+      const iree_hal_semaphore_list_t wait_semaphore_list,
+      const iree_hal_semaphore_list_t signal_semaphore_list,
+      const void* source_buffer, iree_host_size_t source_offset,
+      iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset,
+      iree_device_size_t length, iree_hal_update_flags_t flags);
+
   iree_status_t(IREE_API_PTR* queue_copy)(
       iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
       const iree_hal_semaphore_list_t wait_semaphore_list,
@@ -634,6 +667,14 @@
     iree_device_size_t length, const void* pattern,
     iree_host_size_t pattern_length, iree_hal_fill_flags_t flags);
 
+IREE_API_EXPORT iree_status_t iree_hal_device_queue_emulated_update(
+    iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
+    const iree_hal_semaphore_list_t wait_semaphore_list,
+    const iree_hal_semaphore_list_t signal_semaphore_list,
+    const void* source_buffer, iree_host_size_t source_offset,
+    iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset,
+    iree_device_size_t length, iree_hal_update_flags_t flags);
+
 IREE_API_EXPORT iree_status_t iree_hal_device_queue_emulated_copy(
     iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity,
     const iree_hal_semaphore_list_t wait_semaphore_list,
diff --git a/runtime/src/iree/hal/drivers/cuda/cuda_device.c b/runtime/src/iree/hal/drivers/cuda/cuda_device.c
index 1669510..2ed014b 100644
--- a/runtime/src/iree/hal/drivers/cuda/cuda_device.c
+++ b/runtime/src/iree/hal/drivers/cuda/cuda_device.c
@@ -1130,6 +1130,7 @@
     .queue_alloca = iree_hal_cuda_device_queue_alloca,
     .queue_dealloca = iree_hal_cuda_device_queue_dealloca,
     .queue_fill = iree_hal_device_queue_emulated_fill,
+    .queue_update = iree_hal_device_queue_emulated_update,
     .queue_copy = iree_hal_device_queue_emulated_copy,
     .queue_read = iree_hal_cuda_device_queue_read,
     .queue_write = iree_hal_cuda_device_queue_write,
diff --git a/runtime/src/iree/hal/drivers/hip/hip_device.c b/runtime/src/iree/hal/drivers/hip/hip_device.c
index d065d2c..7f42e8d 100644
--- a/runtime/src/iree/hal/drivers/hip/hip_device.c
+++ b/runtime/src/iree/hal/drivers/hip/hip_device.c
@@ -1127,6 +1127,7 @@
     .queue_alloca = iree_hal_hip_device_queue_alloca,
     .queue_dealloca = iree_hal_hip_device_queue_dealloca,
     .queue_fill = iree_hal_device_queue_emulated_fill,
+    .queue_update = iree_hal_device_queue_emulated_update,
     .queue_copy = iree_hal_device_queue_emulated_copy,
     .queue_read = iree_hal_hip_device_queue_read,
     .queue_write = iree_hal_hip_device_queue_write,
diff --git a/runtime/src/iree/hal/drivers/local_sync/sync_device.c b/runtime/src/iree/hal/drivers/local_sync/sync_device.c
index a543d55..7283e58 100644
--- a/runtime/src/iree/hal/drivers/local_sync/sync_device.c
+++ b/runtime/src/iree/hal/drivers/local_sync/sync_device.c
@@ -504,6 +504,7 @@
     .queue_alloca = iree_hal_sync_device_queue_alloca,
     .queue_dealloca = iree_hal_sync_device_queue_dealloca,
     .queue_fill = iree_hal_device_queue_emulated_fill,
+    .queue_update = iree_hal_device_queue_emulated_update,
     .queue_copy = iree_hal_device_queue_emulated_copy,
     .queue_read = iree_hal_sync_device_queue_read,
     .queue_write = iree_hal_sync_device_queue_write,
diff --git a/runtime/src/iree/hal/drivers/local_task/task_device.c b/runtime/src/iree/hal/drivers/local_task/task_device.c
index d52b08a..8aa0925 100644
--- a/runtime/src/iree/hal/drivers/local_task/task_device.c
+++ b/runtime/src/iree/hal/drivers/local_task/task_device.c
@@ -539,6 +539,7 @@
     .queue_alloca = iree_hal_task_device_queue_alloca,
     .queue_dealloca = iree_hal_task_device_queue_dealloca,
     .queue_fill = iree_hal_device_queue_emulated_fill,
+    .queue_update = iree_hal_device_queue_emulated_update,
     .queue_copy = iree_hal_device_queue_emulated_copy,
     .queue_read = iree_hal_task_device_queue_read,
     .queue_write = iree_hal_task_device_queue_write,
diff --git a/runtime/src/iree/hal/drivers/metal/metal_device.m b/runtime/src/iree/hal/drivers/metal/metal_device.m
index 593abd1..4f8b4fd 100644
--- a/runtime/src/iree/hal/drivers/metal/metal_device.m
+++ b/runtime/src/iree/hal/drivers/metal/metal_device.m
@@ -618,6 +618,7 @@
     .queue_alloca = iree_hal_metal_device_queue_alloca,
     .queue_dealloca = iree_hal_metal_device_queue_dealloca,
     .queue_fill = iree_hal_device_queue_emulated_fill,
+    .queue_update = iree_hal_device_queue_emulated_update,
     .queue_copy = iree_hal_device_queue_emulated_copy,
     .queue_read = iree_hal_metal_device_queue_read,
     .queue_write = iree_hal_metal_device_queue_write,
diff --git a/runtime/src/iree/hal/drivers/null/device.c b/runtime/src/iree/hal/drivers/null/device.c
index 935d9a0..1195364 100644
--- a/runtime/src/iree/hal/drivers/null/device.c
+++ b/runtime/src/iree/hal/drivers/null/device.c
@@ -372,6 +372,26 @@
       target_buffer, target_offset, length, pattern, pattern_length, flags);
 }
 
+static iree_status_t iree_hal_null_device_queue_update(
+    iree_hal_device_t* base_device, iree_hal_queue_affinity_t queue_affinity,
+    const iree_hal_semaphore_list_t wait_semaphore_list,
+    const iree_hal_semaphore_list_t signal_semaphore_list,
+    const void* source_buffer, iree_host_size_t source_offset,
+    iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset,
+    iree_device_size_t length, iree_hal_update_flags_t flags) {
+  // TODO(null): if a native queue update operation is available use that
+  // instead. The emulated update creates a command buffer and executes it and
+  // it's best if the extra recording/upload/allocation time can be avoided.
+  // Since command buffers have a limited capacity for embedded data the
+  // emulated version may need to allocate buffers, split the update into
+  // multiple commands, or commit other sins a native implementation would be
+  // able to avoid.
+  return iree_hal_device_queue_emulated_update(
+      base_device, queue_affinity, wait_semaphore_list, signal_semaphore_list,
+      source_buffer, source_offset, target_buffer, target_offset, length,
+      flags);
+}
+
 static iree_status_t iree_hal_null_device_queue_copy(
     iree_hal_device_t* base_device, iree_hal_queue_affinity_t queue_affinity,
     const iree_hal_semaphore_list_t wait_semaphore_list,
@@ -580,6 +600,7 @@
     .queue_alloca = iree_hal_null_device_queue_alloca,
     .queue_dealloca = iree_hal_null_device_queue_dealloca,
     .queue_fill = iree_hal_null_device_queue_fill,
+    .queue_update = iree_hal_null_device_queue_update,
     .queue_copy = iree_hal_null_device_queue_copy,
     .queue_read = iree_hal_null_device_queue_read,
     .queue_write = iree_hal_null_device_queue_write,
diff --git a/runtime/src/iree/hal/drivers/vulkan/vulkan_device.cc b/runtime/src/iree/hal/drivers/vulkan/vulkan_device.cc
index 2edd31d..6db27bc 100644
--- a/runtime/src/iree/hal/drivers/vulkan/vulkan_device.cc
+++ b/runtime/src/iree/hal/drivers/vulkan/vulkan_device.cc
@@ -1893,6 +1893,7 @@
     /*.queue_alloca=*/iree_hal_vulkan_device_queue_alloca,
     /*.queue_dealloca=*/iree_hal_vulkan_device_queue_dealloca,
     /*.queue_fill=*/iree_hal_device_queue_emulated_fill,
+    /*.queue_update=*/iree_hal_device_queue_emulated_update,
     /*.queue_copy=*/iree_hal_device_queue_emulated_copy,
     /*.queue_read=*/iree_hal_vulkan_device_queue_read,
     /*.queue_write=*/iree_hal_vulkan_device_queue_write,