diff --git a/tests/entrypoints/cohere/test_protocol.py b/tests/entrypoints/cohere/test_protocol.py
index 3e21c1f9ec..5391686c69 100644
--- a/tests/entrypoints/cohere/test_protocol.py
+++ b/tests/entrypoints/cohere/test_protocol.py
@@ -16,6 +16,7 @@ types plus a few local additions:
 import pytest
 from pydantic import ValidationError
 
+import vllm.envs as envs
 from vllm.entrypoints.cohere.protocol import (
     AssistantMessageResponse,
     CitationEndEvent,
@@ -94,6 +95,24 @@ class TestCohereChatV2Request:
                 tool_choice="ANY",  # not REQUIRED/NONE
             )
 
+    def test_stop_sequences_bounded_by_env_limit(self):
+        messages = [{"role": "user", "content": "hi"}]
+        limit = envs.VLLM_MAX_STOP_STRINGS
+
+        req = CohereChatV2Request(
+            model="m",
+            messages=messages,
+            stop_sequences=[f"s{i}" for i in range(limit)],
+        )
+        assert len(req.stop_sequences) == limit
+
+        with pytest.raises(ValidationError):
+            CohereChatV2Request(
+                model="m",
+                messages=messages,
+                stop_sequences=[f"s{i}" for i in range(limit + 1)],
+            )
+
 
 # ======================================================================
 # Usage / Logprobs
diff --git a/vllm/entrypoints/cohere/protocol.py b/vllm/entrypoints/cohere/protocol.py
index 8f3b1f4289..483216a699 100644
--- a/vllm/entrypoints/cohere/protocol.py
+++ b/vllm/entrypoints/cohere/protocol.py
@@ -25,7 +25,7 @@ See https://docs.cohere.com/reference/chat for the upstream spec.
 
 from __future__ import annotations
 
-from typing import Any, Literal
+from typing import Annotated, Any, Literal
 
 from cohere import types as _sdk
 from cohere.types import (
@@ -46,6 +46,7 @@ from cohere.types import (
 )
 from pydantic import BaseModel, Field, field_validator
 
+import vllm.envs as envs
 from vllm.exceptions import VLLMValidationError
 
 # Re-export the SDK wire-format types alongside our local extensions so
@@ -152,7 +153,9 @@ class CohereChatV2Request(BaseModel):
     response_format: ResponseFormatV2 | None = None
     safety_mode: ChatRequestSafetyMode | None = None
     max_tokens: int | None = None
-    stop_sequences: list[str] | None = None
+    stop_sequences: (
+        Annotated[list[str], Field(max_length=envs.VLLM_MAX_STOP_STRINGS)] | None
+    ) = None
 
     # Sampling
     temperature: float | None = None