diff --git a/vllm/inputs/registry.py b/vllm/inputs/registry.py
index 6331a70b469a..dc3236508348 100644
--- a/vllm/inputs/registry.py
+++ b/vllm/inputs/registry.py
@@ -8,10 +8,10 @@
 from transformers import BatchFeature, PretrainedConfig, ProcessorMixin
 from typing_extensions import TypeVar
 
-from vllm.jsontree import JSONTree, json_map_leaves
 from vllm.logger import init_logger
 from vllm.transformers_utils.processor import cached_processor_from_config
 from vllm.utils import get_allowed_kwarg_only_overrides
+from vllm.utils.jsontree import JSONTree, json_map_leaves
 
 if TYPE_CHECKING:
     from vllm.config import ModelConfig
diff --git a/vllm/model_executor/models/aya_vision.py b/vllm/model_executor/models/aya_vision.py
index b476a4f918bc..5cd74bbba482 100644
--- a/vllm/model_executor/models/aya_vision.py
+++ b/vllm/model_executor/models/aya_vision.py
@@ -16,7 +16,6 @@
     get_optimal_tiled_canvas)
 
 from vllm.config import VllmConfig
-from vllm.jsontree import json_map_leaves
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import MultiModalDataDict, MultiModalKwargs
@@ -29,6 +28,7 @@
                                         PromptUpdateDetails)
 from vllm.multimodal.profiling import BaseDummyInputsBuilder
 from vllm.sequence import IntermediateTensors
+from vllm.utils.jsontree import json_map_leaves
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
 
 from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsPP
diff --git a/vllm/model_executor/models/llava.py b/vllm/model_executor/models/llava.py
index 708ca9899522..89d2817b57e0 100644
--- a/vllm/model_executor/models/llava.py
+++ b/vllm/model_executor/models/llava.py
@@ -16,7 +16,6 @@
 
 from vllm.config import VllmConfig
 from vllm.inputs import InputProcessingContext
-from vllm.jsontree import json_map_leaves
 from vllm.model_executor.layers.activation import get_act_fn
 from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                                RowParallelLinear)
@@ -33,6 +32,7 @@
                                         PromptUpdateDetails)
 from vllm.multimodal.profiling import BaseDummyInputsBuilder
 from vllm.sequence import IntermediateTensors
+from vllm.utils.jsontree import json_map_leaves
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
 
 from .clip import CLIPVisionModel
diff --git a/vllm/model_executor/models/minimax_vl_01.py b/vllm/model_executor/models/minimax_vl_01.py
index 62a7d37ec9d3..8107c6e8a04a 100644
--- a/vllm/model_executor/models/minimax_vl_01.py
+++ b/vllm/model_executor/models/minimax_vl_01.py
@@ -8,7 +8,6 @@
 from transformers import BatchFeature, PretrainedConfig
 
 from vllm.config import VllmConfig
-from vllm.jsontree import json_map_leaves
 from vllm.model_executor.layers.activation import get_act_fn
 from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                                RowParallelLinear)
@@ -17,6 +16,7 @@
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import MultiModalFieldConfig
 from vllm.sequence import IntermediateTensors
+from vllm.utils.jsontree import json_map_leaves
 
 from .clip import CLIPVisionModel
 from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsPP
diff --git a/vllm/model_executor/models/tarsier.py b/vllm/model_executor/models/tarsier.py
index 70cf5e95a54e..c8709d866b1e 100644
--- a/vllm/model_executor/models/tarsier.py
+++ b/vllm/model_executor/models/tarsier.py
@@ -18,7 +18,6 @@
 
 from vllm.config import VllmConfig
 from vllm.inputs import InputProcessingContext
-from vllm.jsontree import json_map_leaves
 from vllm.model_executor.layers.activation import get_act_fn
 from vllm.model_executor.layers.linear import (ColumnParallelLinear,
                                                RowParallelLinear)
@@ -34,6 +33,7 @@
                                         PromptReplacement, PromptUpdate)
 from vllm.multimodal.profiling import BaseDummyInputsBuilder
 from vllm.sequence import IntermediateTensors
+from vllm.utils.jsontree import json_map_leaves
 
 from .clip import CLIPVisionModel
 from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsPP
diff --git a/vllm/multimodal/cache.py b/vllm/multimodal/cache.py
index 262b22e554b9..6074a4d54f22 100644
--- a/vllm/multimodal/cache.py
+++ b/vllm/multimodal/cache.py
@@ -7,9 +7,9 @@
 
 import torch
 
-from vllm.jsontree import json_map_leaves, json_reduce_leaves
 from vllm.logger import init_logger
 from vllm.utils import GiB_bytes, LRUCache
+from vllm.utils.jsontree import json_map_leaves, json_reduce_leaves
 
 from .inputs import MultiModalKwargs, MultiModalKwargsItem, NestedTensors
 
diff --git a/vllm/multimodal/inputs.py b/vllm/multimodal/inputs.py
index 18aae35c6fd4..6d4bcef3206c 100644
--- a/vllm/multimodal/inputs.py
+++ b/vllm/multimodal/inputs.py
@@ -13,8 +13,8 @@
 import numpy as np
 from typing_extensions import NotRequired, TypeAlias
 
-from vllm.jsontree import JSONTree, json_map_leaves
 from vllm.utils import LazyLoader, full_groupby, is_list_of
+from vllm.utils.jsontree import JSONTree, json_map_leaves
 
 if TYPE_CHECKING:
     import torch
diff --git a/vllm/jsontree.py b/vllm/utils/jsontree.py
similarity index 100%
rename from vllm/jsontree.py
rename to vllm/utils/jsontree.py