diff --git a/tests/samplers/test_sampler.py b/tests/samplers/test_sampler.py
index 6924aba11576..90340f8cff03 100644
--- a/tests/samplers/test_sampler.py
+++ b/tests/samplers/test_sampler.py
@@ -478,7 +478,7 @@ def test_sampler_mixed(seed: int, device: str):
             sampling_params = SamplingParams(
                 temperature=random.random() + 0.1,
                 top_p=min(random.random() + 0.1, 1),
-                top_k=random.randint(0, 10) or -1,
+                top_k=random.randint(0, 10),
                 n=n,
                 presence_penalty=random.randint(0, 1),
             )
diff --git a/vllm/entrypoints/openai/protocol.py b/vllm/entrypoints/openai/protocol.py
index 40e477f03194..1aa400741340 100644
--- a/vllm/entrypoints/openai/protocol.py
+++ b/vllm/entrypoints/openai/protocol.py
@@ -409,7 +409,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
         "repetition_penalty": 1.0,
         "temperature": 1.0,
         "top_p": 1.0,
-        "top_k": -1,
+        "top_k": 0,
         "min_p": 0.0,
     }
 
@@ -853,7 +853,7 @@ class CompletionRequest(OpenAIBaseModel):
         "repetition_penalty": 1.0,
         "temperature": 1.0,
         "top_p": 1.0,
-        "top_k": -1,
+        "top_k": 0,
         "min_p": 0.0,
     }
 
@@ -1679,7 +1679,7 @@ class TranscriptionRequest(OpenAIBaseModel):
         "repetition_penalty": 1.0,
         "temperature": 1.0,
         "top_p": 1.0,
-        "top_k": -1,
+        "top_k": 0,
         "min_p": 0.0,
     }
 
diff --git a/vllm/model_executor/sampling_metadata.py b/vllm/model_executor/sampling_metadata.py
index d76c75d9e6ce..888ca3e5009e 100644
--- a/vllm/model_executor/sampling_metadata.py
+++ b/vllm/model_executor/sampling_metadata.py
@@ -416,7 +416,7 @@ class SamplingTensors:
 
             # k should not be greater than the vocab size.
             top_k = min(sampling_params.top_k, vocab_size)
-            top_k = vocab_size if top_k == -1 else top_k
+            top_k = vocab_size if top_k < 1 else top_k
             if temperature < _SAMPLING_EPS:
                 # NOTE: Zero temperature means deterministic sampling
                 # (i.e., greedy sampling or beam search).
diff --git a/vllm/sampling_params.py b/vllm/sampling_params.py
index affc5c64b941..dc38daa388ce 100644
--- a/vllm/sampling_params.py
+++ b/vllm/sampling_params.py
@@ -149,7 +149,7 @@ class SamplingParams(
         top_p: Float that controls the cumulative probability of the top tokens
             to consider. Must be in (0, 1]. Set to 1 to consider all tokens.
         top_k: Integer that controls the number of top tokens to consider. Set
-            to -1 to consider all tokens.
+            to 0 (or -1) to consider all tokens.
         min_p: Float that represents the minimum probability for a token to be
             considered, relative to the probability of the most likely token.
             Must be in [0, 1]. Set to 0 to disable this.
@@ -209,7 +209,7 @@ class SamplingParams(
     repetition_penalty: float = 1.0
     temperature: float = 1.0
     top_p: float = 1.0
-    top_k: int = -1
+    top_k: int = 0
     min_p: float = 0.0
     seed: Optional[int] = None
     stop: Optional[Union[str, list[str]]] = None
@@ -256,7 +256,7 @@ class SamplingParams(
         repetition_penalty: Optional[float] = 1.0,
         temperature: Optional[float] = 1.0,
         top_p: Optional[float] = 1.0,
-        top_k: int = -1,
+        top_k: int = 0,
         min_p: float = 0.0,
         seed: Optional[int] = None,
         stop: Optional[Union[str, list[str]]] = None,
@@ -376,7 +376,7 @@ class SamplingParams(
         if self.temperature < _SAMPLING_EPS:
             # Zero temperature means greedy sampling.
             self.top_p = 1.0
-            self.top_k = -1
+            self.top_k = 0
             self.min_p = 0.0
             self._verify_greedy_sampling()
 
@@ -404,8 +404,9 @@ class SamplingParams(
                 f"temperature must be non-negative, got {self.temperature}.")
         if not 0.0 < self.top_p <= 1.0:
             raise ValueError(f"top_p must be in (0, 1], got {self.top_p}.")
-        if self.top_k < -1 or self.top_k == 0:
-            raise ValueError(f"top_k must be -1 (disable), or at least 1, "
+        # quietly accept -1 as disabled, but prefer 0
+        if self.top_k < -1:
+            raise ValueError(f"top_k must be 0 (disable), or at least 1, "
                              f"got {self.top_k}.")
         if not isinstance(self.top_k, int):
             raise TypeError(
diff --git a/vllm/worker/neuron_model_runner.py b/vllm/worker/neuron_model_runner.py
index c80b69e78dc0..e97adf757cc1 100644
--- a/vllm/worker/neuron_model_runner.py
+++ b/vllm/worker/neuron_model_runner.py
@@ -348,7 +348,7 @@ class NeuronModelRunner(ModelRunnerBase[ModelInputForNeuron]):
         if temperature == 0.0:
             # Enable greedy sampling on zero temperature
             return (1, 1.0, 1.0)
-        if top_k < 0 or top_k > self._MAX_NEURON_SAMPLING_TOP_K:
+        if top_k < 1 or top_k > self._MAX_NEURON_SAMPLING_TOP_K:
             top_k = self._MAX_NEURON_SAMPLING_TOP_K
 
         return (top_k, top_p, temperature)
diff --git a/vllm/worker/tpu_model_runner.py b/vllm/worker/tpu_model_runner.py
index 53541a2579ed..e0cca9072745 100644
--- a/vllm/worker/tpu_model_runner.py
+++ b/vllm/worker/tpu_model_runner.py
@@ -525,7 +525,7 @@ class TPUModelRunner(ModelRunnerBase[ModelInputForTPU]):
                     "Top-p sampling is currently disabled for the TPU backend "
                     "due to performance issues.")
             p.append(sampling_params.top_p)
-            if sampling_params.top_k != -1:
+            if sampling_params.top_k > 0:
                 raise NotImplementedError(
                     "Top-k sampling is currently disabled for the TPU backend "
                     "due to performance issues.")