diff --git a/vllm/multimodal/registry.py b/vllm/multimodal/registry.py
index 5b1890de0..e29c7bda0 100644
--- a/vllm/multimodal/registry.py
+++ b/vllm/multimodal/registry.py
@@ -254,7 +254,7 @@ class MultiModalRegistry:
         observability_config: "ObservabilityConfig | None" = None,
         tokenizer: TokenizerLike | None = None,
     ) -> InputProcessingContext:
-        if tokenizer is None and not model_config.skip_tokenizer_init:
+        if tokenizer is None:
             tokenizer = cached_tokenizer_from_config(model_config)
 
         return InputProcessingContext(
diff --git a/vllm/v1/engine/async_llm.py b/vllm/v1/engine/async_llm.py
index 454e20ad1..855758d21 100644
--- a/vllm/v1/engine/async_llm.py
+++ b/vllm/v1/engine/async_llm.py
@@ -106,10 +106,7 @@ class AsyncLLM(EngineClient):
                 "enabling logging without default stat loggers."
             )
 
-        if self.model_config.skip_tokenizer_init:
-            tokenizer = None
-        else:
-            tokenizer = cached_tokenizer_from_config(self.model_config)
+        tokenizer = cached_tokenizer_from_config(self.model_config)
 
         self.input_processor = InputProcessor(self.vllm_config, tokenizer)
         self.io_processor = get_io_processor(
diff --git a/vllm/v1/engine/llm_engine.py b/vllm/v1/engine/llm_engine.py
index c02143c72..78eeb70f1 100644
--- a/vllm/v1/engine/llm_engine.py
+++ b/vllm/v1/engine/llm_engine.py
@@ -84,10 +84,7 @@ class LLMEngine:
             self.dp_group = None
         self.should_execute_dummy_batch = False
 
-        if self.model_config.skip_tokenizer_init:
-            tokenizer = None
-        else:
-            tokenizer = cached_tokenizer_from_config(self.model_config)
+        tokenizer = cached_tokenizer_from_config(self.model_config)
 
         self.input_processor = InputProcessor(self.vllm_config, tokenizer)
         self.io_processor = get_io_processor(