diff --git a/vllm/reasoning/abs_reasoning_parsers.py b/vllm/reasoning/abs_reasoning_parsers.py
index 496eaaf3f..83c3e6b90 100644
--- a/vllm/reasoning/abs_reasoning_parsers.py
+++ b/vllm/reasoning/abs_reasoning_parsers.py
@@ -4,7 +4,7 @@
 import importlib
 import os
 from abc import abstractmethod
-from collections.abc import Callable, Sequence
+from collections.abc import Callable, Iterable, Sequence
 from functools import cached_property
 from typing import TYPE_CHECKING, Any
 
@@ -68,7 +68,7 @@ class ReasoningParser:
         """
 
     def is_reasoning_end_streaming(
-        self, input_ids: Sequence[int], delta_ids: Sequence[int]
+        self, input_ids: Sequence[int], delta_ids: Iterable[int]
     ) -> bool:
         """
         Check if the reasoning content ends in the input_ids on a
diff --git a/vllm/reasoning/basic_parsers.py b/vllm/reasoning/basic_parsers.py
index c066032fb..5b1c0111c 100644
--- a/vllm/reasoning/basic_parsers.py
+++ b/vllm/reasoning/basic_parsers.py
@@ -2,7 +2,8 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 from abc import abstractmethod
-from collections.abc import Sequence
+from collections.abc import Iterable, Sequence
+from itertools import islice
 from typing import TYPE_CHECKING, Any
 
 from vllm.entrypoints.openai.engine.protocol import DeltaMessage
@@ -77,7 +78,7 @@ class BaseThinkingReasoningParser(ReasoningParser):
         return False
 
     def is_reasoning_end_streaming(
-        self, input_ids: Sequence[int], delta_ids: Sequence[int]
+        self, input_ids: Sequence[int], delta_ids: Iterable[int]
     ) -> bool:
         end_token_id = self.end_token_id
         return end_token_id in delta_ids
@@ -86,7 +87,7 @@ class BaseThinkingReasoningParser(ReasoningParser):
         """
         Extract the content after the end tokens
         """
-        if self.end_token_id not in input_ids[:-1]:
+        if self.end_token_id not in islice(input_ids, 0, max(0, len(input_ids) - 1)):
             return []
         else:
             return input_ids[input_ids.index(self.end_token_id) + 1 :]
diff --git a/vllm/reasoning/deepseek_v3_reasoning_parser.py b/vllm/reasoning/deepseek_v3_reasoning_parser.py
index e40f22590..c2efe6500 100644
--- a/vllm/reasoning/deepseek_v3_reasoning_parser.py
+++ b/vllm/reasoning/deepseek_v3_reasoning_parser.py
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
-from collections.abc import Sequence
+from collections.abc import Iterable, Sequence
 
 from transformers import PreTrainedTokenizerBase
 
@@ -41,7 +41,7 @@ class DeepSeekV3ReasoningParser(ReasoningParser):
         return self._parser.is_reasoning_end(input_ids)
 
     def is_reasoning_end_streaming(
-        self, input_ids: Sequence[int], delta_ids: Sequence[int]
+        self, input_ids: Sequence[int], delta_ids: Iterable[int]
     ) -> bool:
         return self._parser.is_reasoning_end_streaming(input_ids, delta_ids)
 
diff --git a/vllm/reasoning/identity_reasoning_parser.py b/vllm/reasoning/identity_reasoning_parser.py
index e1106362d..3c76901a3 100644
--- a/vllm/reasoning/identity_reasoning_parser.py
+++ b/vllm/reasoning/identity_reasoning_parser.py
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
-from collections.abc import Sequence
+from collections.abc import Iterable, Sequence
 
 from transformers import PreTrainedTokenizerBase
 
@@ -36,7 +36,7 @@ class IdentityReasoningParser(ReasoningParser):
         return True
 
     def is_reasoning_end_streaming(
-        self, input_ids: Sequence[int], delta_ids: Sequence[int]
+        self, input_ids: Sequence[int], delta_ids: Iterable[int]
     ) -> bool:
         return True
 
diff --git a/vllm/reasoning/mistral_reasoning_parser.py b/vllm/reasoning/mistral_reasoning_parser.py
index d73474626..c085ba4e4 100644
--- a/vllm/reasoning/mistral_reasoning_parser.py
+++ b/vllm/reasoning/mistral_reasoning_parser.py
@@ -69,7 +69,7 @@ class MistralReasoningParser(BaseThinkingReasoningParser):
     def is_reasoning_end(self, input_ids: Sequence[int]) -> bool:
         has_eot_token = False
 
-        for id in input_ids[::-1]:
+        for id in reversed(input_ids):
             if id == self.start_token_id:
                 # Reasoning ends only if a BOT token is found before a EOT token.
                 return has_eot_token
diff --git a/vllm/reasoning/step3_reasoning_parser.py b/vllm/reasoning/step3_reasoning_parser.py
index 4758246ac..d932ba8b6 100644
--- a/vllm/reasoning/step3_reasoning_parser.py
+++ b/vllm/reasoning/step3_reasoning_parser.py
@@ -1,7 +1,8 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
-from collections.abc import Sequence
+from collections.abc import Iterable, Sequence
+from itertools import islice
 
 import regex as re
 from transformers import PreTrainedTokenizerBase
@@ -104,13 +105,15 @@ class Step3ReasoningParser(ReasoningParser):
         return self.think_end_token_id in input_ids
 
     def is_reasoning_end_streaming(
-        self, input_ids: Sequence[int], delta_ids: Sequence[int]
+        self, input_ids: Sequence[int], delta_ids: Iterable[int]
     ) -> bool:
         end_token_id = self.think_end_token_id
         return end_token_id in delta_ids
 
     def extract_content_ids(self, input_ids: list[int]) -> list[int]:
-        if self.think_end_token_id not in input_ids[:-1]:
+        if self.think_end_token_id not in islice(
+            input_ids, 0, max(0, len(input_ids) - 1)
+        ):
             return []
         else:
             return input_ids[input_ids.index(self.think_end_token_id) + 1 :]
diff --git a/vllm/reasoning/step3p5_reasoning_parser.py b/vllm/reasoning/step3p5_reasoning_parser.py
index b93f55142..af9aa4b41 100644
--- a/vllm/reasoning/step3p5_reasoning_parser.py
+++ b/vllm/reasoning/step3p5_reasoning_parser.py
@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
-from collections.abc import Sequence
+from collections.abc import Iterable, Sequence
 
 from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
@@ -51,7 +51,7 @@ class Step3p5ReasoningParser(BaseThinkingReasoningParser):
         return self.end_offset < 1
 
     def is_reasoning_end_streaming(
-        self, input_ids: Sequence[int], delta_ids: Sequence[int]
+        self, input_ids: Sequence[int], delta_ids: Iterable[int]
     ) -> bool:
         if self.end_token_id in input_ids and self.end_offset > 0:
             self.end_offset -= 1
diff --git a/vllm/v1/structured_output/__init__.py b/vllm/v1/structured_output/__init__.py
index 921bee6a6..cd17a21d9 100644
--- a/vllm/v1/structured_output/__init__.py
+++ b/vllm/v1/structured_output/__init__.py
@@ -324,8 +324,11 @@ class StructuredOutputManager:
         # Check if reasoning ends in *this* step
         delta_from = request.num_computed_tokens - request.num_output_placeholders
         all_token_ids = request.all_token_ids
+        start = (
+            delta_from if delta_from >= 0 else max(len(all_token_ids) + delta_from, 0)
+        )
         if self.reasoner.is_reasoning_end_streaming(
-            all_token_ids, all_token_ids[delta_from:]
+            all_token_ids, itertools.islice(all_token_ids, start, None)
         ):
             # Reasoning just ended, so we shouldn't advance til
             # next pass