From 0d9a576fb07002492679e6946668df6fc7757307 Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Thu, 19 Jun 2025 11:13:13 +0200
Subject: [PATCH 01/19] Initial commit

---
 haystack/components/generators/chat/openai.py |  5 ++
 haystack/components/generators/utils.py       | 12 ++--
 haystack/dataclasses/__init__.py              |  2 +
 haystack/dataclasses/streaming_chunk.py       | 39 ++++++++++++
 test/dataclasses/test_streaming_chunk.py      | 61 ++++++++++++++++++-
 5 files changed, 112 insertions(+), 7 deletions(-)

diff --git a/haystack/components/generators/chat/openai.py b/haystack/components/generators/chat/openai.py
index fda06e6ebd..03592828c3 100644
--- a/haystack/components/generators/chat/openai.py
+++ b/haystack/components/generators/chat/openai.py
@@ -530,10 +530,13 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
                 component_info=component_info,
                 # Index is None since it's only set to an int when a content block is present
                 index=None,
+                finish_reason=None,
                 meta={
                     "model": chunk.model,
                     "received_at": datetime.now().isoformat(),
                     "usage": _serialize_usage(chunk.usage),
+                    # NOTE: finish_reason in meta is deprecated, use the dedicated finish_reason field
+                    "finish_reason": None,
                 },
             )
         ]
@@ -557,6 +560,7 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
                     arguments=function.arguments if function and function.arguments else None,
                 ),
                 start=function.name is not None if function else False,
+                finish_reason=choice.finish_reason,
                 meta={
                     "model": chunk.model,
                     "index": choice.index,
@@ -579,6 +583,7 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
         # The first chunk is always a start message chunk that only contains role information, so if we reach here
         # and previous_chunks is length 1 then this is the start of text content.
         start=len(previous_chunks) == 1,
+        finish_reason=choice.finish_reason,
         meta={
             "model": chunk.model,
             "index": choice.index,
diff --git a/haystack/components/generators/utils.py b/haystack/components/generators/utils.py
index e66b13786e..bd003b0797 100644
--- a/haystack/components/generators/utils.py
+++ b/haystack/components/generators/utils.py
@@ -58,7 +58,7 @@ def print_streaming_chunk(chunk: StreamingChunk) -> None:
 
     # End of LLM assistant message so we add two new lines
     # This ensures spacing between multiple LLM messages (e.g. Agent) or multiple Tool Call Results
-    if chunk.meta.get("finish_reason") is not None:
+    if chunk.finish_reason is not None:
         print("\n\n", flush=True, end="")
 
 
@@ -112,9 +112,13 @@ def _convert_streaming_chunks_to_chat_message(chunks: List[StreamingChunk]) -> C
             )
 
     # finish_reason can appear in different places so we look for the last one
-    finish_reasons = [
-        chunk.meta.get("finish_reason") for chunk in chunks if chunk.meta.get("finish_reason") is not None
-    ]
+    # First check the dedicated finish_reason field, then fall back to meta for backward compatibility
+    finish_reasons = []
+    for chunk in chunks:
+        if chunk.finish_reason is not None:
+            finish_reasons.append(chunk.finish_reason)
+        elif chunk.meta.get("finish_reason") is not None:
+            finish_reasons.append(chunk.meta.get("finish_reason"))
     finish_reason = finish_reasons[-1] if finish_reasons else None
 
     meta = {
diff --git a/haystack/dataclasses/__init__.py b/haystack/dataclasses/__init__.py
index 1356df131a..442a2c0004 100644
--- a/haystack/dataclasses/__init__.py
+++ b/haystack/dataclasses/__init__.py
@@ -22,6 +22,7 @@
         "SyncStreamingCallbackT",
         "ToolCallDelta",
         "select_streaming_callback",
+        "FinishReason",
     ],
 }
 
@@ -40,6 +41,7 @@
     from .state import State as State
     from .streaming_chunk import AsyncStreamingCallbackT as AsyncStreamingCallbackT
     from .streaming_chunk import ComponentInfo as ComponentInfo
+    from .streaming_chunk import FinishReason as FinishReason
     from .streaming_chunk import StreamingCallbackT as StreamingCallbackT
     from .streaming_chunk import StreamingChunk as StreamingChunk
     from .streaming_chunk import SyncStreamingCallbackT as SyncStreamingCallbackT
diff --git a/haystack/dataclasses/streaming_chunk.py b/haystack/dataclasses/streaming_chunk.py
index d01a61bc05..f98fbfb97f 100644
--- a/haystack/dataclasses/streaming_chunk.py
+++ b/haystack/dataclasses/streaming_chunk.py
@@ -2,6 +2,7 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
+import warnings
 from dataclasses import dataclass, field
 from typing import Any, Awaitable, Callable, Dict, Literal, Optional, Union, overload
 
@@ -9,6 +10,9 @@
 from haystack.dataclasses.chat_message import ToolCallResult
 from haystack.utils.asynchronous import is_callable_async_compatible
 
+# Type alias for standard finish_reason values following OpenAI's convention
+FinishReason = Literal["stop", "length", "tool_calls", "content_filter"]
+
 
 @dataclass
 class ToolCallDelta:
@@ -68,12 +72,16 @@ class StreamingChunk:
 
     :param content: The content of the message chunk as a string.
     :param meta: A dictionary containing metadata related to the message chunk.
+        NOTE: The 'finish_reason' field in meta is deprecated and will be removed in a future release.
+        Use the dedicated 'finish_reason' field instead.
     :param component_info: A `ComponentInfo` object containing information about the component that generated the chunk,
         such as the component name and type.
     :param index: An optional integer index representing which content block this chunk belongs to.
     :param tool_call: An optional ToolCallDelta object representing a tool call associated with the message chunk.
     :param tool_call_result: An optional ToolCallResult object representing the result of a tool call.
     :param start: A boolean indicating whether this chunk marks the start of a content block.
+    :param finish_reason: An optional string indicating the reason the generation finished.
+        Standard values follow OpenAI's convention: "stop", "length", "tool_calls", "content_filter".
     """
 
     content: str
@@ -83,6 +91,7 @@ class StreamingChunk:
     tool_call: Optional[ToolCallDelta] = field(default=None)
     tool_call_result: Optional[ToolCallResult] = field(default=None)
     start: bool = field(default=False)
+    finish_reason: Optional[Union[FinishReason, str]] = field(default=None)
 
     def __post_init__(self):
         fields_set = sum(bool(x) for x in (self.content, self.tool_call, self.tool_call_result))
@@ -97,6 +106,10 @@ def __post_init__(self):
         if (self.tool_call or self.tool_call_result) and self.index is None:
             raise ValueError("If `tool_call`, or `tool_call_result` is set, `index` must also be set.")
 
+        # Convert meta to _DeprecationWarningDict to show warnings when accessing deprecated fields
+        if not isinstance(self.meta, _DeprecationWarningDict):
+            self.meta = _DeprecationWarningDict(self.meta)
+
 
 SyncStreamingCallbackT = Callable[[StreamingChunk], None]
 AsyncStreamingCallbackT = Callable[[StreamingChunk], Awaitable[None]]
@@ -148,3 +161,29 @@ def select_streaming_callback(
             raise ValueError("The runtime callback cannot be a coroutine.")
 
     return runtime_callback or init_callback
+
+
+class _DeprecationWarningDict(dict):
+    """
+    A dictionary subclass that issues deprecation warnings when accessing 'finish_reason'.
+    """
+
+    def __getitem__(self, key):
+        if key == "finish_reason":
+            warnings.warn(
+                "Accessing 'finish_reason' from StreamingChunk.meta is deprecated and will be removed in a future "
+                "release. Use StreamingChunk.finish_reason instead.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+        return super().__getitem__(key)
+
+    def get(self, key, default=None):
+        if key == "finish_reason":
+            warnings.warn(
+                "Accessing 'finish_reason' from StreamingChunk.meta is deprecated and will be removed in a future "
+                "release. Use StreamingChunk.finish_reason instead.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+        return super().get(key, default)
diff --git a/test/dataclasses/test_streaming_chunk.py b/test/dataclasses/test_streaming_chunk.py
index 695d155483..56d1479a2c 100644
--- a/test/dataclasses/test_streaming_chunk.py
+++ b/test/dataclasses/test_streaming_chunk.py
@@ -4,9 +4,8 @@
 
 import pytest
 
-from haystack.dataclasses import StreamingChunk, ComponentInfo, ToolCallDelta, ToolCallResult, ToolCall
-from haystack import component
-from haystack import Pipeline
+from haystack import Pipeline, component
+from haystack.dataclasses import ComponentInfo, FinishReason, StreamingChunk, ToolCall, ToolCallDelta, ToolCallResult
 
 
 @component
@@ -101,3 +100,59 @@ def test_tool_call_delta():
 def test_tool_call_delta_with_missing_fields():
     with pytest.raises(ValueError):
         _ = ToolCallDelta(id="123")
+
+
+def test_create_chunk_with_finish_reason():
+    """Test creating a chunk with the new finish_reason field."""
+    chunk = StreamingChunk(content="Test content", finish_reason="stop")
+
+    assert chunk.content == "Test content"
+    assert chunk.finish_reason == "stop"
+    assert chunk.meta == {}
+
+
+def test_create_chunk_with_finish_reason_and_meta():
+    """Test creating a chunk with both finish_reason field and meta."""
+    chunk = StreamingChunk(
+        content="Test content", finish_reason="stop", meta={"model": "gpt-4", "usage": {"tokens": 10}}
+    )
+
+    assert chunk.content == "Test content"
+    assert chunk.finish_reason == "stop"
+    assert chunk.meta["model"] == "gpt-4"
+    assert chunk.meta["usage"]["tokens"] == 10
+
+
+def test_finish_reason_deprecation_warning():
+    """Test that accessing finish_reason via meta shows deprecation warning."""
+    import warnings
+
+    chunk = StreamingChunk(content="Test content", meta={"finish_reason": "length"})
+
+    with warnings.catch_warnings(record=True) as w:
+        warnings.simplefilter("always")
+        result = chunk.meta.get("finish_reason")
+
+        assert len(w) == 1
+        assert issubclass(w[0].category, DeprecationWarning)
+        assert "finish_reason" in str(w[0].message)
+        assert "deprecated" in str(w[0].message)
+        assert result == "length"
+
+
+def test_finish_reason_openai_standard_values():
+    """Test that the finish_reason field accepts OpenAI standard values."""
+    openai_values = ["stop", "length", "tool_calls", "content_filter"]
+
+    for value in openai_values:
+        chunk = StreamingChunk(content="Test content", finish_reason=value)
+        assert chunk.finish_reason == value
+
+
+def test_finish_reason_custom_values():
+    """Test that custom finish_reason values still work (for migration compatibility)."""
+    custom_values = ["custom_stop", "provider_specific", "unknown"]
+
+    for value in custom_values:
+        chunk = StreamingChunk(content="Test content", finish_reason=value)
+        assert chunk.finish_reason == value

From ee41abf5e0c2c05d7b3c82724ce3629e61d3f8d6 Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Thu, 19 Jun 2025 11:22:30 +0200
Subject: [PATCH 02/19] Update deprecation version

---
 haystack/dataclasses/streaming_chunk.py  | 12 +++++++-----
 test/dataclasses/test_streaming_chunk.py | 15 ++++-----------
 2 files changed, 11 insertions(+), 16 deletions(-)

diff --git a/haystack/dataclasses/streaming_chunk.py b/haystack/dataclasses/streaming_chunk.py
index f98fbfb97f..a34d41fcaf 100644
--- a/haystack/dataclasses/streaming_chunk.py
+++ b/haystack/dataclasses/streaming_chunk.py
@@ -72,7 +72,7 @@ class StreamingChunk:
 
     :param content: The content of the message chunk as a string.
     :param meta: A dictionary containing metadata related to the message chunk.
-        NOTE: The 'finish_reason' field in meta is deprecated and will be removed in a future release.
+        NOTE: The 'finish_reason' field in meta is deprecated and will be removed in Haystack 2.17.
         Use the dedicated 'finish_reason' field instead.
     :param component_info: A `ComponentInfo` object containing information about the component that generated the chunk,
         such as the component name and type.
@@ -91,6 +91,8 @@ class StreamingChunk:
     tool_call: Optional[ToolCallDelta] = field(default=None)
     tool_call_result: Optional[ToolCallResult] = field(default=None)
     start: bool = field(default=False)
+    # NOTE: We allow for the finish_reason to be a string during migration phase
+    # we will change finish_reason to be a FinishReason in a future release
     finish_reason: Optional[Union[FinishReason, str]] = field(default=None)
 
     def __post_init__(self):
@@ -171,8 +173,8 @@ class _DeprecationWarningDict(dict):
     def __getitem__(self, key):
         if key == "finish_reason":
             warnings.warn(
-                "Accessing 'finish_reason' from StreamingChunk.meta is deprecated and will be removed in a future "
-                "release. Use StreamingChunk.finish_reason instead.",
+                "Accessing 'finish_reason' from StreamingChunk.meta is deprecated and will be removed in "
+                "Haystack 2.17. Use StreamingChunk.finish_reason instead.",
                 DeprecationWarning,
                 stacklevel=2,
             )
@@ -181,8 +183,8 @@ def __getitem__(self, key):
     def get(self, key, default=None):
         if key == "finish_reason":
             warnings.warn(
-                "Accessing 'finish_reason' from StreamingChunk.meta is deprecated and will be removed in a future "
-                "release. Use StreamingChunk.finish_reason instead.",
+                "Accessing 'finish_reason' from StreamingChunk.meta is deprecated and will be removed in "
+                "Haystack 2.17. Use StreamingChunk.finish_reason instead.",
                 DeprecationWarning,
                 stacklevel=2,
             )
diff --git a/test/dataclasses/test_streaming_chunk.py b/test/dataclasses/test_streaming_chunk.py
index 56d1479a2c..c1435d965b 100644
--- a/test/dataclasses/test_streaming_chunk.py
+++ b/test/dataclasses/test_streaming_chunk.py
@@ -4,8 +4,9 @@
 
 import pytest
 
-from haystack import Pipeline, component
-from haystack.dataclasses import ComponentInfo, FinishReason, StreamingChunk, ToolCall, ToolCallDelta, ToolCallResult
+from haystack.dataclasses import StreamingChunk, ComponentInfo, ToolCallDelta, ToolCallResult, ToolCall, FinishReason
+from haystack import component
+from haystack import Pipeline
 
 
 @component
@@ -137,18 +138,10 @@ def test_finish_reason_deprecation_warning():
         assert issubclass(w[0].category, DeprecationWarning)
         assert "finish_reason" in str(w[0].message)
         assert "deprecated" in str(w[0].message)
+        assert "Haystack 2.17" in str(w[0].message)
         assert result == "length"
 
 
-def test_finish_reason_openai_standard_values():
-    """Test that the finish_reason field accepts OpenAI standard values."""
-    openai_values = ["stop", "length", "tool_calls", "content_filter"]
-
-    for value in openai_values:
-        chunk = StreamingChunk(content="Test content", finish_reason=value)
-        assert chunk.finish_reason == value
-
-
 def test_finish_reason_custom_values():
     """Test that custom finish_reason values still work (for migration compatibility)."""
     custom_values = ["custom_stop", "provider_specific", "unknown"]

From e59c10aee65e7dac156a16c353d5a3f1e66059ad Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Thu, 19 Jun 2025 11:29:54 +0200
Subject: [PATCH 03/19] Improve comment

---
 haystack/components/generators/utils.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/haystack/components/generators/utils.py b/haystack/components/generators/utils.py
index bd003b0797..740377945e 100644
--- a/haystack/components/generators/utils.py
+++ b/haystack/components/generators/utils.py
@@ -113,6 +113,8 @@ def _convert_streaming_chunks_to_chat_message(chunks: List[StreamingChunk]) -> C
 
     # finish_reason can appear in different places so we look for the last one
     # First check the dedicated finish_reason field, then fall back to meta for backward compatibility
+    # NOTE: This fallback is required during migration period until all chat generators
+    # are updated to populate the dedicated finish_reason field.
     finish_reasons = []
     for chunk in chunks:
         if chunk.finish_reason is not None:

From 41b0d9ad5e460fafce9ba35c1b6619a2c882a247 Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Thu, 19 Jun 2025 15:10:14 +0200
Subject: [PATCH 04/19] Minor simplification

---
 haystack/components/generators/utils.py | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/haystack/components/generators/utils.py b/haystack/components/generators/utils.py
index 740377945e..7169ad11f8 100644
--- a/haystack/components/generators/utils.py
+++ b/haystack/components/generators/utils.py
@@ -114,13 +114,12 @@ def _convert_streaming_chunks_to_chat_message(chunks: List[StreamingChunk]) -> C
     # finish_reason can appear in different places so we look for the last one
     # First check the dedicated finish_reason field, then fall back to meta for backward compatibility
     # NOTE: This fallback is required during migration period until all chat generators
-    # are updated to populate the dedicated finish_reason field.
-    finish_reasons = []
-    for chunk in chunks:
-        if chunk.finish_reason is not None:
-            finish_reasons.append(chunk.finish_reason)
-        elif chunk.meta.get("finish_reason") is not None:
-            finish_reasons.append(chunk.meta.get("finish_reason"))
+    # are updated to populate the dedicated StreamingChunk.finish_reason field.
+    finish_reasons = [
+        chunk.finish_reason or chunk.meta.get("finish_reason")
+        for chunk in chunks
+        if chunk.finish_reason or chunk.meta.get("finish_reason")
+    ]
     finish_reason = finish_reasons[-1] if finish_reasons else None
 
     meta = {

From 632780a5ec8147eafa996bfd790a2195bff09953 Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Thu, 19 Jun 2025 15:16:36 +0200
Subject: [PATCH 05/19] Add reno note

---
 ...nish-reason-field-streaming-chunk-89828ec09c6e6385.yaml | 7 +++++++
 1 file changed, 7 insertions(+)
 create mode 100644 releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml

diff --git a/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml b/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml
new file mode 100644
index 0000000000..ac002a480b
--- /dev/null
+++ b/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml
@@ -0,0 +1,7 @@
+---
+features:
+  - |
+    Added dedicated `finish_reason` field to `StreamingChunk` class to improve type safety and enable sophisticated streaming UI logic. The field follows OpenAI's convention with standard values: "stop", "length", "tool_calls", "content_filter".
+deprecations:
+  - |
+    Accessing `finish_reason` from `StreamingChunk.meta` is now deprecated and will be removed in Haystack 2.17. Use the dedicated `StreamingChunk.finish_reason` field instead. During the migration period, both approaches work but accessing via meta shows deprecation warnings.

From 29ee6ba54dcefdbc555c4b6cf062020717cb309f Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 10:45:19 +0200
Subject: [PATCH 06/19] Remove deprecation warning

---
 haystack/components/generators/chat/openai.py |  2 +-
 haystack/components/generators/utils.py       |  4 +--
 haystack/dataclasses/streaming_chunk.py       | 34 ++-----------------
 test/dataclasses/test_streaming_chunk.py      | 29 ++++++++++------
 4 files changed, 24 insertions(+), 45 deletions(-)

diff --git a/haystack/components/generators/chat/openai.py b/haystack/components/generators/chat/openai.py
index 03592828c3..87f83cb1d2 100644
--- a/haystack/components/generators/chat/openai.py
+++ b/haystack/components/generators/chat/openai.py
@@ -535,7 +535,7 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
                     "model": chunk.model,
                     "received_at": datetime.now().isoformat(),
                     "usage": _serialize_usage(chunk.usage),
-                    # NOTE: finish_reason in meta is deprecated, use the dedicated finish_reason field
+                    # NOTE: Both finish_reason field and meta["finish_reason"] are supported for flexibility
                     "finish_reason": None,
                 },
             )
diff --git a/haystack/components/generators/utils.py b/haystack/components/generators/utils.py
index 7169ad11f8..e3addabf36 100644
--- a/haystack/components/generators/utils.py
+++ b/haystack/components/generators/utils.py
@@ -113,8 +113,8 @@ def _convert_streaming_chunks_to_chat_message(chunks: List[StreamingChunk]) -> C
 
     # finish_reason can appear in different places so we look for the last one
     # First check the dedicated finish_reason field, then fall back to meta for backward compatibility
-    # NOTE: This fallback is required during migration period until all chat generators
-    # are updated to populate the dedicated StreamingChunk.finish_reason field.
+    # NOTE: Support both finish_reason field and meta fallback for maximum flexibility.
+    # Users can choose either approach based on their preferences.
     finish_reasons = [
         chunk.finish_reason or chunk.meta.get("finish_reason")
         for chunk in chunks
diff --git a/haystack/dataclasses/streaming_chunk.py b/haystack/dataclasses/streaming_chunk.py
index a34d41fcaf..2c2726983d 100644
--- a/haystack/dataclasses/streaming_chunk.py
+++ b/haystack/dataclasses/streaming_chunk.py
@@ -72,8 +72,8 @@ class StreamingChunk:
 
     :param content: The content of the message chunk as a string.
     :param meta: A dictionary containing metadata related to the message chunk.
-        NOTE: The 'finish_reason' field in meta is deprecated and will be removed in Haystack 2.17.
-        Use the dedicated 'finish_reason' field instead.
+        NOTE: Both 'finish_reason' field and meta['finish_reason'] are supported for flexibility.
+        The dedicated field provides better type safety and IDE support.
     :param component_info: A `ComponentInfo` object containing information about the component that generated the chunk,
         such as the component name and type.
     :param index: An optional integer index representing which content block this chunk belongs to.
@@ -108,10 +108,6 @@ def __post_init__(self):
         if (self.tool_call or self.tool_call_result) and self.index is None:
             raise ValueError("If `tool_call`, or `tool_call_result` is set, `index` must also be set.")
 
-        # Convert meta to _DeprecationWarningDict to show warnings when accessing deprecated fields
-        if not isinstance(self.meta, _DeprecationWarningDict):
-            self.meta = _DeprecationWarningDict(self.meta)
-
 
 SyncStreamingCallbackT = Callable[[StreamingChunk], None]
 AsyncStreamingCallbackT = Callable[[StreamingChunk], Awaitable[None]]
@@ -163,29 +159,3 @@ def select_streaming_callback(
             raise ValueError("The runtime callback cannot be a coroutine.")
 
     return runtime_callback or init_callback
-
-
-class _DeprecationWarningDict(dict):
-    """
-    A dictionary subclass that issues deprecation warnings when accessing 'finish_reason'.
-    """
-
-    def __getitem__(self, key):
-        if key == "finish_reason":
-            warnings.warn(
-                "Accessing 'finish_reason' from StreamingChunk.meta is deprecated and will be removed in "
-                "Haystack 2.17. Use StreamingChunk.finish_reason instead.",
-                DeprecationWarning,
-                stacklevel=2,
-            )
-        return super().__getitem__(key)
-
-    def get(self, key, default=None):
-        if key == "finish_reason":
-            warnings.warn(
-                "Accessing 'finish_reason' from StreamingChunk.meta is deprecated and will be removed in "
-                "Haystack 2.17. Use StreamingChunk.finish_reason instead.",
-                DeprecationWarning,
-                stacklevel=2,
-            )
-        return super().get(key, default)
diff --git a/test/dataclasses/test_streaming_chunk.py b/test/dataclasses/test_streaming_chunk.py
index c1435d965b..b7b3817cf9 100644
--- a/test/dataclasses/test_streaming_chunk.py
+++ b/test/dataclasses/test_streaming_chunk.py
@@ -124,22 +124,31 @@ def test_create_chunk_with_finish_reason_and_meta():
     assert chunk.meta["usage"]["tokens"] == 10
 
 
-def test_finish_reason_deprecation_warning():
-    """Test that accessing finish_reason via meta shows deprecation warning."""
+def test_dual_finish_reason_support():
+    """Test that both finish_reason approaches work without warnings."""
     import warnings
 
+    # Test that accessing finish_reason via meta does NOT show warnings
     chunk = StreamingChunk(content="Test content", meta={"finish_reason": "length"})
 
     with warnings.catch_warnings(record=True) as w:
         warnings.simplefilter("always")
-        result = chunk.meta.get("finish_reason")
-
-        assert len(w) == 1
-        assert issubclass(w[0].category, DeprecationWarning)
-        assert "finish_reason" in str(w[0].message)
-        assert "deprecated" in str(w[0].message)
-        assert "Haystack 2.17" in str(w[0].message)
-        assert result == "length"
+        result_meta = chunk.meta.get("finish_reason")
+        result_direct = chunk.meta["finish_reason"]
+
+        # No warnings should be raised
+        assert len(w) == 0
+        assert result_meta == "length"
+        assert result_direct == "length"
+
+    # Test that both approaches can be used simultaneously
+    chunk_dual = StreamingChunk(
+        content="Test content", finish_reason="stop", meta={"finish_reason": "length", "model": "gpt-4"}
+    )
+
+    assert chunk_dual.finish_reason == "stop"
+    assert chunk_dual.meta["finish_reason"] == "length"
+    assert chunk_dual.meta.get("finish_reason") == "length"
 
 
 def test_finish_reason_custom_values():

From 4504bbf7643fa60fc616d381ce87a6549ec51fe5 Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 11:51:13 +0200
Subject: [PATCH 07/19] Remove fallback in
 haystack/components/generators/utils.py

---
 haystack/components/generators/utils.py | 9 +--------
 1 file changed, 1 insertion(+), 8 deletions(-)

diff --git a/haystack/components/generators/utils.py b/haystack/components/generators/utils.py
index e3addabf36..b66dd9d7d8 100644
--- a/haystack/components/generators/utils.py
+++ b/haystack/components/generators/utils.py
@@ -112,14 +112,7 @@ def _convert_streaming_chunks_to_chat_message(chunks: List[StreamingChunk]) -> C
             )
 
     # finish_reason can appear in different places so we look for the last one
-    # First check the dedicated finish_reason field, then fall back to meta for backward compatibility
-    # NOTE: Support both finish_reason field and meta fallback for maximum flexibility.
-    # Users can choose either approach based on their preferences.
-    finish_reasons = [
-        chunk.finish_reason or chunk.meta.get("finish_reason")
-        for chunk in chunks
-        if chunk.finish_reason or chunk.meta.get("finish_reason")
-    ]
+    finish_reasons = [chunk.finish_reason for chunk in chunks if chunk.finish_reason]
     finish_reason = finish_reasons[-1] if finish_reasons else None
 
     meta = {

From a329d87ef2ca76648a275bda38becd8f563ad84f Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 11:53:55 +0200
Subject: [PATCH 08/19] FinishReason alphabetical import

---
 haystack/dataclasses/__init__.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/haystack/dataclasses/__init__.py b/haystack/dataclasses/__init__.py
index 442a2c0004..91071227f5 100644
--- a/haystack/dataclasses/__init__.py
+++ b/haystack/dataclasses/__init__.py
@@ -17,12 +17,12 @@
     "streaming_chunk": [
         "AsyncStreamingCallbackT",
         "ComponentInfo",
+        "FinishReason",
         "StreamingCallbackT",
         "StreamingChunk",
         "SyncStreamingCallbackT",
         "ToolCallDelta",
         "select_streaming_callback",
-        "FinishReason",
     ],
 }
 

From 7aa2f9c94404ccda2762c509a0327c189d6a392c Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 12:03:27 +0200
Subject: [PATCH 09/19] Add tool_call_results finish reason, adapt codebase

---
 haystack/components/tools/tool_invoker.py  | 12 ++++-
 haystack/dataclasses/streaming_chunk.py    |  3 +-
 test/components/tools/test_tool_invoker.py | 52 ++++++++++++++++++++++
 test/dataclasses/test_streaming_chunk.py   | 18 ++++++++
 4 files changed, 82 insertions(+), 3 deletions(-)

diff --git a/haystack/components/tools/tool_invoker.py b/haystack/components/tools/tool_invoker.py
index 7c7dd9faed..47f2f0eca5 100644
--- a/haystack/components/tools/tool_invoker.py
+++ b/haystack/components/tools/tool_invoker.py
@@ -553,7 +553,11 @@ def run(
 
         # We stream one more chunk that contains a finish_reason if tool_messages were generated
         if len(tool_messages) > 0 and streaming_callback is not None:
-            streaming_callback(StreamingChunk(content="", meta={"finish_reason": "tool_call_results"}))
+            streaming_callback(
+                StreamingChunk(
+                    content="", finish_reason="tool_call_results", meta={"finish_reason": "tool_call_results"}
+                )
+            )
 
         return {"tool_messages": tool_messages, "state": state}
 
@@ -685,7 +689,11 @@ async def run_async(
 
         # We stream one more chunk that contains a finish_reason if tool_messages were generated
         if len(tool_messages) > 0 and streaming_callback is not None:
-            await streaming_callback(StreamingChunk(content="", meta={"finish_reason": "tool_call_results"}))
+            await streaming_callback(
+                StreamingChunk(
+                    content="", finish_reason="tool_call_results", meta={"finish_reason": "tool_call_results"}
+                )
+            )
 
         return {"tool_messages": tool_messages, "state": state}
 
diff --git a/haystack/dataclasses/streaming_chunk.py b/haystack/dataclasses/streaming_chunk.py
index 2c2726983d..f10b4b8ead 100644
--- a/haystack/dataclasses/streaming_chunk.py
+++ b/haystack/dataclasses/streaming_chunk.py
@@ -11,7 +11,8 @@
 from haystack.utils.asynchronous import is_callable_async_compatible
 
 # Type alias for standard finish_reason values following OpenAI's convention
-FinishReason = Literal["stop", "length", "tool_calls", "content_filter"]
+# plus Haystack-specific values
+FinishReason = Literal["stop", "length", "tool_calls", "content_filter", "tool_call_results"]
 
 
 @dataclass
diff --git a/test/components/tools/test_tool_invoker.py b/test/components/tools/test_tool_invoker.py
index a21f5ae31a..9e14fcac56 100644
--- a/test/components/tools/test_tool_invoker.py
+++ b/test/components/tools/test_tool_invoker.py
@@ -203,6 +203,28 @@ def streaming_callback(chunk: StreamingChunk) -> None:
         assert tool_call_result.origin == tool_call
         assert not tool_call_result.error
 
+    def test_run_with_streaming_callback_finish_reason(self, invoker):
+        streaming_chunks = []
+
+        def streaming_callback(chunk: StreamingChunk) -> None:
+            streaming_chunks.append(chunk)
+
+        tool_call = ToolCall(tool_name="weather_tool", arguments={"location": "Berlin"})
+        message = ChatMessage.from_assistant(tool_calls=[tool_call])
+
+        result = invoker.run(messages=[message], streaming_callback=streaming_callback)
+        assert "tool_messages" in result
+        assert len(result["tool_messages"]) == 1
+
+        # Check that we received streaming chunks
+        assert len(streaming_chunks) >= 2  # At least one for tool result and one for finish reason
+
+        # The last chunk should have finish_reason set to "tool_call_results"
+        final_chunk = streaming_chunks[-1]
+        assert final_chunk.finish_reason == "tool_call_results"
+        assert final_chunk.meta["finish_reason"] == "tool_call_results"
+        assert final_chunk.content == ""
+
     @pytest.mark.asyncio
     async def test_run_async_with_streaming_callback(self, thread_executor, weather_tool):
         streaming_callback_called = False
@@ -245,6 +267,36 @@ async def streaming_callback(chunk: StreamingChunk) -> None:
         # check we called the streaming callback
         assert streaming_callback_called
 
+    @pytest.mark.asyncio
+    async def test_run_async_with_streaming_callback_finish_reason(self, thread_executor, weather_tool):
+        streaming_chunks = []
+
+        async def streaming_callback(chunk: StreamingChunk) -> None:
+            streaming_chunks.append(chunk)
+
+        tool_invoker = ToolInvoker(
+            tools=[weather_tool],
+            raise_on_failure=True,
+            convert_result_to_json_string=False,
+            async_executor=thread_executor,
+        )
+
+        tool_call = ToolCall(tool_name="weather_tool", arguments={"location": "Berlin"})
+        message = ChatMessage.from_assistant(tool_calls=[tool_call])
+
+        result = await tool_invoker.run_async(messages=[message], streaming_callback=streaming_callback)
+        assert "tool_messages" in result
+        assert len(result["tool_messages"]) == 1
+
+        # Check that we received streaming chunks
+        assert len(streaming_chunks) >= 2  # At least one for tool result and one for finish reason
+
+        # The last chunk should have finish_reason set to "tool_call_results"
+        final_chunk = streaming_chunks[-1]
+        assert final_chunk.finish_reason == "tool_call_results"
+        assert final_chunk.meta["finish_reason"] == "tool_call_results"
+        assert final_chunk.content == ""
+
     def test_run_with_toolset(self, tool_set):
         tool_invoker = ToolInvoker(tools=tool_set, raise_on_failure=True, convert_result_to_json_string=False)
         tool_call = ToolCall(tool_name="addition_tool", arguments={"num1": 5, "num2": 3})
diff --git a/test/dataclasses/test_streaming_chunk.py b/test/dataclasses/test_streaming_chunk.py
index b7b3817cf9..7a2f5c1fe7 100644
--- a/test/dataclasses/test_streaming_chunk.py
+++ b/test/dataclasses/test_streaming_chunk.py
@@ -158,3 +158,21 @@ def test_finish_reason_custom_values():
     for value in custom_values:
         chunk = StreamingChunk(content="Test content", finish_reason=value)
         assert chunk.finish_reason == value
+
+
+def test_finish_reason_standard_values():
+    """Test all standard finish_reason values including the new Haystack-specific ones."""
+    standard_values = ["stop", "length", "tool_calls", "content_filter", "tool_call_results"]
+
+    for value in standard_values:
+        chunk = StreamingChunk(content="Test content", finish_reason=value)
+        assert chunk.finish_reason == value
+
+
+def test_finish_reason_tool_call_results():
+    """Test specifically the new tool_call_results finish reason."""
+    chunk = StreamingChunk(content="", finish_reason="tool_call_results", meta={"finish_reason": "tool_call_results"})
+
+    assert chunk.finish_reason == "tool_call_results"
+    assert chunk.meta["finish_reason"] == "tool_call_results"
+    assert chunk.content == ""

From ba143a0510b323439a0e5e82656518cf76a330b9 Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 12:09:11 +0200
Subject: [PATCH 10/19] Define finish_reason to be Optional[FinishReason]

---
 haystack/dataclasses/streaming_chunk.py  | 9 ++++-----
 test/components/generators/test_utils.py | 1 +
 test/dataclasses/test_streaming_chunk.py | 9 ---------
 3 files changed, 5 insertions(+), 14 deletions(-)

diff --git a/haystack/dataclasses/streaming_chunk.py b/haystack/dataclasses/streaming_chunk.py
index f10b4b8ead..7d7a45385f 100644
--- a/haystack/dataclasses/streaming_chunk.py
+++ b/haystack/dataclasses/streaming_chunk.py
@@ -81,8 +81,9 @@ class StreamingChunk:
     :param tool_call: An optional ToolCallDelta object representing a tool call associated with the message chunk.
     :param tool_call_result: An optional ToolCallResult object representing the result of a tool call.
     :param start: A boolean indicating whether this chunk marks the start of a content block.
-    :param finish_reason: An optional string indicating the reason the generation finished.
-        Standard values follow OpenAI's convention: "stop", "length", "tool_calls", "content_filter".
+    :param finish_reason: An optional value indicating the reason the generation finished.
+        Standard values follow OpenAI's convention: "stop", "length", "tool_calls", "content_filter",
+        plus Haystack-specific values like "tool_call_results".
     """
 
     content: str
@@ -92,9 +93,7 @@ class StreamingChunk:
     tool_call: Optional[ToolCallDelta] = field(default=None)
     tool_call_result: Optional[ToolCallResult] = field(default=None)
     start: bool = field(default=False)
-    # NOTE: We allow for the finish_reason to be a string during migration phase
-    # we will change finish_reason to be a FinishReason in a future release
-    finish_reason: Optional[Union[FinishReason, str]] = field(default=None)
+    finish_reason: Optional[FinishReason] = field(default=None)
 
     def __post_init__(self):
         fields_set = sum(bool(x) for x in (self.content, self.tool_call, self.tool_call_result))
diff --git a/test/components/generators/test_utils.py b/test/components/generators/test_utils.py
index cc6d6edfdf..5dc2f0cee6 100644
--- a/test/components/generators/test_utils.py
+++ b/test/components/generators/test_utils.py
@@ -262,6 +262,7 @@ def test_convert_streaming_chunks_to_chat_message_tool_calls_in_any_chunk():
                 "received_at": "2025-02-19T16:02:55.948772",
             },
             component_info=ComponentInfo(name="test", type="test"),
+            finish_reason="tool_calls",
         ),
         StreamingChunk(
             content="",
diff --git a/test/dataclasses/test_streaming_chunk.py b/test/dataclasses/test_streaming_chunk.py
index 7a2f5c1fe7..268d222b52 100644
--- a/test/dataclasses/test_streaming_chunk.py
+++ b/test/dataclasses/test_streaming_chunk.py
@@ -151,15 +151,6 @@ def test_dual_finish_reason_support():
     assert chunk_dual.meta.get("finish_reason") == "length"
 
 
-def test_finish_reason_custom_values():
-    """Test that custom finish_reason values still work (for migration compatibility)."""
-    custom_values = ["custom_stop", "provider_specific", "unknown"]
-
-    for value in custom_values:
-        chunk = StreamingChunk(content="Test content", finish_reason=value)
-        assert chunk.finish_reason == value
-
-
 def test_finish_reason_standard_values():
     """Test all standard finish_reason values including the new Haystack-specific ones."""
     standard_values = ["stop", "length", "tool_calls", "content_filter", "tool_call_results"]

From 88f65da06837c9ea197fa33db5bd58998074497b Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 12:21:25 +0200
Subject: [PATCH 11/19] Add StreamingChunk finish_reason in HF generators

---
 haystack/components/generators/chat/hugging_face_api.py | 1 +
 haystack/components/generators/hugging_face_api.py      | 7 ++++++-
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/haystack/components/generators/chat/hugging_face_api.py b/haystack/components/generators/chat/hugging_face_api.py
index a9f12e7f6c..a79d281d18 100644
--- a/haystack/components/generators/chat/hugging_face_api.py
+++ b/haystack/components/generators/chat/hugging_face_api.py
@@ -141,6 +141,7 @@ def _convert_chat_completion_stream_output_to_streaming_chunk(
         index=0 if choice.finish_reason is None else None,
         # start is True at the very beginning since first chunk contains role information + first part of the answer.
         start=len(previous_chunks) == 0,
+        finish_reason=choice.finish_reason,
     )
     return stream_chunk
 
diff --git a/haystack/components/generators/hugging_face_api.py b/haystack/components/generators/hugging_face_api.py
index 89fb659bed..375cef8d12 100644
--- a/haystack/components/generators/hugging_face_api.py
+++ b/haystack/components/generators/hugging_face_api.py
@@ -242,7 +242,12 @@ def _stream_and_build_response(
                 first_chunk_time = datetime.now().isoformat()
 
             stream_chunk = StreamingChunk(
-                content=token.text, meta=chunk_metadata, component_info=component_info, index=0, start=len(chunks) == 0
+                content=token.text,
+                meta=chunk_metadata,
+                component_info=component_info,
+                index=0,
+                start=len(chunks) == 0,
+                finish_reason=chunk_metadata.get("finish_reason"),
             )
             chunks.append(stream_chunk)
             streaming_callback(stream_chunk)

From 816fc8b9e0e04c09c1bf9564b1600dee71dd18fa Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 12:21:59 +0200
Subject: [PATCH 12/19] Update reno note

---
 ...finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml b/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml
index ac002a480b..b48e045ff8 100644
--- a/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml
+++ b/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml
@@ -1,7 +1,6 @@
 ---
 features:
   - |
-    Added dedicated `finish_reason` field to `StreamingChunk` class to improve type safety and enable sophisticated streaming UI logic. The field follows OpenAI's convention with standard values: "stop", "length", "tool_calls", "content_filter".
-deprecations:
+    Added dedicated `finish_reason` field to `StreamingChunk` class to improve type safety and enable sophisticated streaming UI logic. The field uses a `FinishReason` type alias with standard values: "stop", "length", "tool_calls", "content_filter", plus Haystack-specific values like "tool_call_results" (used by ToolInvoker to indicate tool execution completion). Both the dedicated field and legacy meta access (`chunk.meta["finish_reason"]`) are permanently supported for maximum flexibility. Users can choose either approach based on their preferences, with the dedicated field providing better type safety and IDE support.
   - |
-    Accessing `finish_reason` from `StreamingChunk.meta` is now deprecated and will be removed in Haystack 2.17. Use the dedicated `StreamingChunk.finish_reason` field instead. During the migration period, both approaches work but accessing via meta shows deprecation warnings.
+    Updated `ToolInvoker` component to use the new `finish_reason` field when streaming tool results. The component now sets `finish_reason="tool_call_results"` in the final streaming chunk to indicate that tool execution has completed, while maintaining backward compatibility by also setting the value in `meta["finish_reason"]`.

From ece472515715116c375feb32c8f1f824cc6c8d28 Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 14:15:37 +0200
Subject: [PATCH 13/19] Repair merge issue

---
 haystack/components/generators/chat/openai.py | 43 +++++++------------
 .../components/generators/chat/test_openai.py | 15 ++++++-
 2 files changed, 29 insertions(+), 29 deletions(-)

diff --git a/haystack/components/generators/chat/openai.py b/haystack/components/generators/chat/openai.py
index 0acfe4e625..1f84cb4da2 100644
--- a/haystack/components/generators/chat/openai.py
+++ b/haystack/components/generators/chat/openai.py
@@ -517,27 +517,23 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
         generated the chunk, such as the component name and type.
 
     :returns:
-        A list of StreamingChunk objects representing the content of the chunk from the OpenAI API.
+        A StreamingChunk object representing the content of the chunk from the OpenAI API.
     """
     # On very first chunk so len(previous_chunks) == 0, the Choices field only provides role info (e.g. "assistant")
     # Choices is empty if include_usage is set to True where the usage information is returned.
     if len(chunk.choices) == 0:
-        return [
-            StreamingChunk(
-                content="",
-                component_info=component_info,
-                # Index is None since it's only set to an int when a content block is present
-                index=None,
-                finish_reason=None,
-                meta={
-                    "model": chunk.model,
-                    "received_at": datetime.now().isoformat(),
-                    "usage": _serialize_usage(chunk.usage),
-                    # NOTE: Both finish_reason field and meta["finish_reason"] are supported for flexibility
-                    "finish_reason": None,
-                },
-            )
-        ]
+        return StreamingChunk(
+            content="",
+            component_info=component_info,
+            # Index is None since it's only set to an int when a content block is present
+            index=None,
+            finish_reason=None,
+            meta={
+                "model": chunk.model,
+                "received_at": datetime.now().isoformat(),
+                "usage": _serialize_usage(chunk.usage),
+            },
+        )
 
     choice: ChunkChoice = chunk.choices[0]
 
@@ -552,17 +548,7 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
                     id=tool_call.id,
                     tool_name=function.name if function else None,
                     arguments=function.arguments if function and function.arguments else None,
-                ),
-                start=function.name is not None if function else False,
-                finish_reason=choice.finish_reason,
-                meta={
-                    "model": chunk.model,
-                    "index": choice.index,
-                    "tool_calls": choice.delta.tool_calls,
-                    "finish_reason": choice.finish_reason,
-                    "received_at": datetime.now().isoformat(),
-                    "usage": _serialize_usage(chunk.usage),
-                },
+                )
             )
         chunk_message = StreamingChunk(
             content=choice.delta.content or "",
@@ -571,6 +557,7 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
             index=tool_calls_deltas[0].index,
             tool_calls=tool_calls_deltas,
             start=tool_calls_deltas[0].tool_name is not None,
+            finish_reason=choice.finish_reason,
             meta={
                 "model": chunk.model,
                 "index": choice.index,
diff --git a/test/components/generators/chat/test_openai.py b/test/components/generators/chat/test_openai.py
index fd4c227e1e..3fdd53bb34 100644
--- a/test/components/generators/chat/test_openai.py
+++ b/test/components/generators/chat/test_openai.py
@@ -986,6 +986,7 @@ def streaming_chunks():
                 "received_at": ANY,
                 "usage": None,
             },
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1007,6 +1008,7 @@ def streaming_chunks():
             index=0,
             tool_calls=[ToolCallDelta(tool_name="weather", id="call_zcvlnVaTeJWRjLAFfYxX69z4", index=0)],
             start=True,
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1020,6 +1022,7 @@ def streaming_chunks():
             },
             index=0,
             tool_calls=[ToolCallDelta(arguments='{"ci', index=0)],
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1033,6 +1036,7 @@ def streaming_chunks():
             },
             index=0,
             tool_calls=[ToolCallDelta(arguments='ty": ', index=0)],
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1046,6 +1050,7 @@ def streaming_chunks():
             },
             index=0,
             tool_calls=[ToolCallDelta(arguments='"Paris', index=0)],
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1059,6 +1064,7 @@ def streaming_chunks():
             },
             index=0,
             tool_calls=[ToolCallDelta(arguments='"}', index=0)],
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1080,6 +1086,7 @@ def streaming_chunks():
             index=1,
             tool_calls=[ToolCallDelta(tool_name="weather", id="call_C88m67V16CrETq6jbNXjdZI9", index=1)],
             start=True,
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1093,6 +1100,7 @@ def streaming_chunks():
             },
             index=1,
             tool_calls=[ToolCallDelta(arguments='{"ci', index=1)],
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1106,6 +1114,7 @@ def streaming_chunks():
             },
             index=1,
             tool_calls=[ToolCallDelta(arguments='ty": ', index=1)],
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1119,6 +1128,7 @@ def streaming_chunks():
             },
             index=1,
             tool_calls=[ToolCallDelta(arguments='"Berli', index=1)],
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1132,6 +1142,7 @@ def streaming_chunks():
             },
             index=1,
             tool_calls=[ToolCallDelta(arguments='n"}', index=1)],
+            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1143,6 +1154,7 @@ def streaming_chunks():
                 "received_at": ANY,
                 "usage": None,
             },
+            finish_reason="tool_calls",
         ),
         StreamingChunk(
             content="",
@@ -1162,6 +1174,7 @@ def streaming_chunks():
                     "prompt_tokens_details": {"audio_tokens": 0, "cached_tokens": 0},
                 },
             },
+            finish_reason=None,
         ),
     ]
 
@@ -1174,7 +1187,7 @@ def test_convert_chat_completion_chunk_to_streaming_chunk(self, chat_completion_
                 chunk=openai_chunk, previous_chunks=previous_chunks
             )
             assert stream_chunk == haystack_chunk
-            previous_chunks.append(openai_chunk)
+            previous_chunks.append(stream_chunk)
 
     def test_handle_stream_response(self, chat_completion_chunks):
         openai_chunks = chat_completion_chunks

From 7db10d7dbb782700c3dd0312d6b0985f4f8785f6 Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 14:21:13 +0200
Subject: [PATCH 14/19] Update tests for finish_reason

---
 test/components/generators/chat/test_hugging_face_api.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/test/components/generators/chat/test_hugging_face_api.py b/test/components/generators/chat/test_hugging_face_api.py
index 6698a743ee..a4db4da2b4 100644
--- a/test/components/generators/chat/test_hugging_face_api.py
+++ b/test/components/generators/chat/test_hugging_face_api.py
@@ -687,6 +687,7 @@ def test_convert_hfapi_tool_calls_invalid_type_arguments(self):
                     },
                     index=0,
                     start=True,
+                    finish_reason=None,
                 ),
                 [],
             ),
@@ -711,6 +712,7 @@ def test_convert_hfapi_tool_calls_invalid_type_arguments(self):
                         "model": "microsoft/Phi-3.5-mini-instruct",
                         "finish_reason": "stop",
                     },
+                    finish_reason="stop",
                 ),
                 [0],
             ),
@@ -730,6 +732,7 @@ def test_convert_hfapi_tool_calls_invalid_type_arguments(self):
                         "model": "microsoft/Phi-3.5-mini-instruct",
                         "usage": {"completion_tokens": 2, "prompt_tokens": 21},
                     },
+                    finish_reason=None,
                 ),
                 [0, 1],
             ),

From 0408d779d0ec66488c40d4bc632d6d3969911a9e Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 15:05:05 +0200
Subject: [PATCH 15/19] Resolve mypy issues

---
 haystack/components/generators/chat/hugging_face_api.py | 4 +++-
 haystack/dataclasses/streaming_chunk.py                 | 2 +-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/haystack/components/generators/chat/hugging_face_api.py b/haystack/components/generators/chat/hugging_face_api.py
index a79d281d18..44af62446a 100644
--- a/haystack/components/generators/chat/hugging_face_api.py
+++ b/haystack/components/generators/chat/hugging_face_api.py
@@ -141,7 +141,9 @@ def _convert_chat_completion_stream_output_to_streaming_chunk(
         index=0 if choice.finish_reason is None else None,
         # start is True at the very beginning since first chunk contains role information + first part of the answer.
         start=len(previous_chunks) == 0,
-        finish_reason=choice.finish_reason,
+        # there is no way to constrain here as many models can return any finish reason
+        # so we need to ignore the type error
+        finish_reason=choice.finish_reason,  # type: ignore[arg-type]
     )
     return stream_chunk
 
diff --git a/haystack/dataclasses/streaming_chunk.py b/haystack/dataclasses/streaming_chunk.py
index 3198b7e460..84d868774f 100644
--- a/haystack/dataclasses/streaming_chunk.py
+++ b/haystack/dataclasses/streaming_chunk.py
@@ -12,7 +12,7 @@
 
 # Type alias for standard finish_reason values following OpenAI's convention
 # plus Haystack-specific values
-FinishReason = Literal["stop", "length", "tool_calls", "content_filter", "tool_call_results"]
+FinishReason = Literal["stop", "length", "tool_calls", "function_call", "content_filter", "tool_call_results"]
 
 
 @dataclass

From bd8153520f29473a79cc904af4909577ceea3cef Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Mon, 23 Jun 2025 15:16:50 +0200
Subject: [PATCH 16/19] Lint issue

---
 haystack/dataclasses/streaming_chunk.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/haystack/dataclasses/streaming_chunk.py b/haystack/dataclasses/streaming_chunk.py
index 84d868774f..642358e839 100644
--- a/haystack/dataclasses/streaming_chunk.py
+++ b/haystack/dataclasses/streaming_chunk.py
@@ -2,7 +2,6 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
-import warnings
 from dataclasses import dataclass, field
 from typing import Any, Awaitable, Callable, Dict, List, Literal, Optional, Union, overload
 

From 59a13494953a838a3e613679260800d21c847294 Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Tue, 24 Jun 2025 10:20:13 +0200
Subject: [PATCH 17/19] Enhance HF finish_reason translation

---
 .../generators/chat/hugging_face_api.py       | 43 +++++++++++++++++--
 1 file changed, 40 insertions(+), 3 deletions(-)

diff --git a/haystack/components/generators/chat/hugging_face_api.py b/haystack/components/generators/chat/hugging_face_api.py
index 44af62446a..d9e69e5023 100644
--- a/haystack/components/generators/chat/hugging_face_api.py
+++ b/haystack/components/generators/chat/hugging_face_api.py
@@ -18,6 +18,7 @@
     ToolCall,
     select_streaming_callback,
 )
+from haystack.dataclasses.streaming_chunk import FinishReason
 from haystack.lazy_imports import LazyImport
 from haystack.tools import (
     Tool,
@@ -41,6 +42,7 @@
         ChatCompletionOutput,
         ChatCompletionOutputToolCall,
         ChatCompletionStreamOutput,
+        ChatCompletionStreamOutputChoice,
         InferenceClient,
     )
 
@@ -110,6 +112,42 @@ def _convert_tools_to_hfapi_tools(
     return hf_tools
 
 
+def _map_hf_finish_reason_to_haystack(choice: "ChatCompletionStreamOutputChoice") -> Optional[FinishReason]:
+    """
+    Map HuggingFace finish reasons to Haystack FinishReason literals.
+
+    Uses the full choice object to detect tool calls and provide accurate mapping.
+
+    HuggingFace finish reasons:
+    - "length": number of generated tokens == `max_new_tokens`
+    - "eos_token": the model generated its end of sequence token
+    - "stop_sequence": the model generated a text included in `stop_sequences`
+
+    Additionally detects tool calls from delta.tool_calls or delta.tool_call_id.
+
+    :param choice: The HuggingFace ChatCompletionStreamOutputChoice object.
+    :returns: The corresponding Haystack FinishReason or None.
+    """
+    if choice.finish_reason is None:
+        return None
+
+    # Check if this choice contains tool call information
+    has_tool_calls = choice.delta.tool_calls is not None or choice.delta.tool_call_id is not None
+
+    # If we detect tool calls, override the finish reason
+    if has_tool_calls:
+        return "tool_calls"
+
+    # Map HuggingFace finish reasons to Haystack standard ones
+    mapping: Dict[str, FinishReason] = {
+        "length": "length",  # Direct match
+        "eos_token": "stop",  # EOS token means natural stop
+        "stop_sequence": "stop",  # Stop sequence means natural stop
+    }
+
+    return mapping.get(choice.finish_reason, "stop")  # Default to "stop" for unknown reasons
+
+
 def _convert_chat_completion_stream_output_to_streaming_chunk(
     chunk: "ChatCompletionStreamOutput",
     previous_chunks: List[StreamingChunk],
@@ -133,6 +171,7 @@ def _convert_chat_completion_stream_output_to_streaming_chunk(
     # the argument is probably allowed for compatibility with OpenAI
     # see https://huggingface.co/docs/huggingface_hub/package_reference/inference_client#huggingface_hub.InferenceClient.chat_completion.n
     choice = chunk.choices[0]
+    mapped_finish_reason = _map_hf_finish_reason_to_haystack(choice)
     stream_chunk = StreamingChunk(
         content=choice.delta.content or "",
         meta={"model": chunk.model, "received_at": datetime.now().isoformat(), "finish_reason": choice.finish_reason},
@@ -141,9 +180,7 @@ def _convert_chat_completion_stream_output_to_streaming_chunk(
         index=0 if choice.finish_reason is None else None,
         # start is True at the very beginning since first chunk contains role information + first part of the answer.
         start=len(previous_chunks) == 0,
-        # there is no way to constrain here as many models can return any finish reason
-        # so we need to ignore the type error
-        finish_reason=choice.finish_reason,  # type: ignore[arg-type]
+        finish_reason=mapped_finish_reason,
     )
     return stream_chunk
 

From 399956cd570603c958bfabc41873b525091f0d1e Mon Sep 17 00:00:00 2001
From: Vladimir Blagojevic <dovlex@gmail.com>
Date: Tue, 24 Jun 2025 10:50:56 +0200
Subject: [PATCH 18/19] Remove irrlevant test

---
 test/dataclasses/test_streaming_chunk.py | 27 ------------------------
 1 file changed, 27 deletions(-)

diff --git a/test/dataclasses/test_streaming_chunk.py b/test/dataclasses/test_streaming_chunk.py
index 3ee702259d..1d53633026 100644
--- a/test/dataclasses/test_streaming_chunk.py
+++ b/test/dataclasses/test_streaming_chunk.py
@@ -125,33 +125,6 @@ def test_create_chunk_with_finish_reason_and_meta():
     assert chunk.meta["usage"]["tokens"] == 10
 
 
-def test_dual_finish_reason_support():
-    """Test that both finish_reason approaches work without warnings."""
-    import warnings
-
-    # Test that accessing finish_reason via meta does NOT show warnings
-    chunk = StreamingChunk(content="Test content", meta={"finish_reason": "length"})
-
-    with warnings.catch_warnings(record=True) as w:
-        warnings.simplefilter("always")
-        result_meta = chunk.meta.get("finish_reason")
-        result_direct = chunk.meta["finish_reason"]
-
-        # No warnings should be raised
-        assert len(w) == 0
-        assert result_meta == "length"
-        assert result_direct == "length"
-
-    # Test that both approaches can be used simultaneously
-    chunk_dual = StreamingChunk(
-        content="Test content", finish_reason="stop", meta={"finish_reason": "length", "model": "gpt-4"}
-    )
-
-    assert chunk_dual.finish_reason == "stop"
-    assert chunk_dual.meta["finish_reason"] == "length"
-    assert chunk_dual.meta.get("finish_reason") == "length"
-
-
 def test_finish_reason_standard_values():
     """Test all standard finish_reason values including the new Haystack-specific ones."""
     standard_values = ["stop", "length", "tool_calls", "content_filter", "tool_call_results"]

From c2d846734783d618bba58f9fa11585e0e8b3b5d4 Mon Sep 17 00:00:00 2001
From: Sebastian Husch Lee <sjrl423@gmail.com>
Date: Wed, 25 Jun 2025 10:56:59 +0200
Subject: [PATCH 19/19] PR comments

---
 .../components/generators/chat/hugging_face_api.py   |  5 +++--
 haystack/components/generators/chat/openai.py        | 12 ++++++++++--
 haystack/components/generators/hugging_face_api.py   | 11 ++++++++++-
 haystack/dataclasses/streaming_chunk.py              |  8 +++-----
 ...eason-field-streaming-chunk-89828ec09c6e6385.yaml |  2 +-
 .../generators/chat/test_hugging_face_api.py         |  2 --
 test/components/generators/chat/test_openai.py       | 12 ------------
 7 files changed, 27 insertions(+), 25 deletions(-)

diff --git a/haystack/components/generators/chat/hugging_face_api.py b/haystack/components/generators/chat/hugging_face_api.py
index d9e69e5023..7a34a5aa3c 100644
--- a/haystack/components/generators/chat/hugging_face_api.py
+++ b/haystack/components/generators/chat/hugging_face_api.py
@@ -118,7 +118,8 @@ def _map_hf_finish_reason_to_haystack(choice: "ChatCompletionStreamOutputChoice"
 
     Uses the full choice object to detect tool calls and provide accurate mapping.
 
-    HuggingFace finish reasons:
+    HuggingFace finish reasons (can be found here https://huggingface.github.io/text-generation-inference/ under
+    FinishReason):
     - "length": number of generated tokens == `max_new_tokens`
     - "eos_token": the model generated its end of sequence token
     - "stop_sequence": the model generated a text included in `stop_sequences`
@@ -171,7 +172,7 @@ def _convert_chat_completion_stream_output_to_streaming_chunk(
     # the argument is probably allowed for compatibility with OpenAI
     # see https://huggingface.co/docs/huggingface_hub/package_reference/inference_client#huggingface_hub.InferenceClient.chat_completion.n
     choice = chunk.choices[0]
-    mapped_finish_reason = _map_hf_finish_reason_to_haystack(choice)
+    mapped_finish_reason = _map_hf_finish_reason_to_haystack(choice) if choice.finish_reason else None
     stream_chunk = StreamingChunk(
         content=choice.delta.content or "",
         meta={"model": chunk.model, "received_at": datetime.now().isoformat(), "finish_reason": choice.finish_reason},
diff --git a/haystack/components/generators/chat/openai.py b/haystack/components/generators/chat/openai.py
index 1f84cb4da2..b9e0988dfb 100644
--- a/haystack/components/generators/chat/openai.py
+++ b/haystack/components/generators/chat/openai.py
@@ -18,6 +18,7 @@
     AsyncStreamingCallbackT,
     ChatMessage,
     ComponentInfo,
+    FinishReason,
     StreamingCallbackT,
     StreamingChunk,
     SyncStreamingCallbackT,
@@ -519,6 +520,13 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
     :returns:
         A StreamingChunk object representing the content of the chunk from the OpenAI API.
     """
+    finish_reason_mapping: Dict[str, FinishReason] = {
+        "stop": "stop",
+        "length": "length",
+        "content_filter": "content_filter",
+        "tool_calls": "tool_calls",
+        "function_call": "tool_calls",
+    }
     # On very first chunk so len(previous_chunks) == 0, the Choices field only provides role info (e.g. "assistant")
     # Choices is empty if include_usage is set to True where the usage information is returned.
     if len(chunk.choices) == 0:
@@ -557,7 +565,7 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
             index=tool_calls_deltas[0].index,
             tool_calls=tool_calls_deltas,
             start=tool_calls_deltas[0].tool_name is not None,
-            finish_reason=choice.finish_reason,
+            finish_reason=finish_reason_mapping.get(choice.finish_reason) if choice.finish_reason else None,
             meta={
                 "model": chunk.model,
                 "index": choice.index,
@@ -586,7 +594,7 @@ def _convert_chat_completion_chunk_to_streaming_chunk(
         # The first chunk is always a start message chunk that only contains role information, so if we reach here
         # and previous_chunks is length 1 then this is the start of text content.
         start=len(previous_chunks) == 1,
-        finish_reason=choice.finish_reason,
+        finish_reason=finish_reason_mapping.get(choice.finish_reason) if choice.finish_reason else None,
         meta={
             "model": chunk.model,
             "index": choice.index,
diff --git a/haystack/components/generators/hugging_face_api.py b/haystack/components/generators/hugging_face_api.py
index 375cef8d12..0da42871ab 100644
--- a/haystack/components/generators/hugging_face_api.py
+++ b/haystack/components/generators/hugging_face_api.py
@@ -9,6 +9,7 @@
 from haystack import component, default_from_dict, default_to_dict
 from haystack.dataclasses import (
     ComponentInfo,
+    FinishReason,
     StreamingCallbackT,
     StreamingChunk,
     SyncStreamingCallbackT,
@@ -241,13 +242,21 @@ def _stream_and_build_response(
             if first_chunk_time is None:
                 first_chunk_time = datetime.now().isoformat()
 
+            mapping: Dict[str, FinishReason] = {
+                "length": "length",  # Direct match
+                "eos_token": "stop",  # EOS token means natural stop
+                "stop_sequence": "stop",  # Stop sequence means natural stop
+            }
+            mapped_finish_reason = (
+                mapping.get(chunk_metadata["finish_reason"], "stop") if chunk_metadata.get("finish_reason") else None
+            )
             stream_chunk = StreamingChunk(
                 content=token.text,
                 meta=chunk_metadata,
                 component_info=component_info,
                 index=0,
                 start=len(chunks) == 0,
-                finish_reason=chunk_metadata.get("finish_reason"),
+                finish_reason=mapped_finish_reason,
             )
             chunks.append(stream_chunk)
             streaming_callback(stream_chunk)
diff --git a/haystack/dataclasses/streaming_chunk.py b/haystack/dataclasses/streaming_chunk.py
index 642358e839..7cd30e466f 100644
--- a/haystack/dataclasses/streaming_chunk.py
+++ b/haystack/dataclasses/streaming_chunk.py
@@ -10,8 +10,8 @@
 from haystack.utils.asynchronous import is_callable_async_compatible
 
 # Type alias for standard finish_reason values following OpenAI's convention
-# plus Haystack-specific values
-FinishReason = Literal["stop", "length", "tool_calls", "function_call", "content_filter", "tool_call_results"]
+# plus Haystack-specific value ("tool_call_results")
+FinishReason = Literal["stop", "length", "tool_calls", "content_filter", "tool_call_results"]
 
 
 @dataclass
@@ -74,8 +74,6 @@ class StreamingChunk:
 
     :param content: The content of the message chunk as a string.
     :param meta: A dictionary containing metadata related to the message chunk.
-        NOTE: Both 'finish_reason' field and meta['finish_reason'] are supported for flexibility.
-        The dedicated field provides better type safety and IDE support.
     :param component_info: A `ComponentInfo` object containing information about the component that generated the chunk,
         such as the component name and type.
     :param index: An optional integer index representing which content block this chunk belongs to.
@@ -85,7 +83,7 @@ class StreamingChunk:
     :param start: A boolean indicating whether this chunk marks the start of a content block.
     :param finish_reason: An optional value indicating the reason the generation finished.
         Standard values follow OpenAI's convention: "stop", "length", "tool_calls", "content_filter",
-        plus Haystack-specific values like "tool_call_results".
+        plus Haystack-specific value "tool_call_results".
     """
 
     content: str
diff --git a/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml b/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml
index b48e045ff8..fdceffe5ee 100644
--- a/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml
+++ b/releasenotes/notes/add-finish-reason-field-streaming-chunk-89828ec09c6e6385.yaml
@@ -1,6 +1,6 @@
 ---
 features:
   - |
-    Added dedicated `finish_reason` field to `StreamingChunk` class to improve type safety and enable sophisticated streaming UI logic. The field uses a `FinishReason` type alias with standard values: "stop", "length", "tool_calls", "content_filter", plus Haystack-specific values like "tool_call_results" (used by ToolInvoker to indicate tool execution completion). Both the dedicated field and legacy meta access (`chunk.meta["finish_reason"]`) are permanently supported for maximum flexibility. Users can choose either approach based on their preferences, with the dedicated field providing better type safety and IDE support.
+    Added dedicated `finish_reason` field to `StreamingChunk` class to improve type safety and enable sophisticated streaming UI logic. The field uses a `FinishReason` type alias with standard values: "stop", "length", "tool_calls", "content_filter", plus Haystack-specific value "tool_call_results" (used by ToolInvoker to indicate tool execution completion).
   - |
     Updated `ToolInvoker` component to use the new `finish_reason` field when streaming tool results. The component now sets `finish_reason="tool_call_results"` in the final streaming chunk to indicate that tool execution has completed, while maintaining backward compatibility by also setting the value in `meta["finish_reason"]`.
diff --git a/test/components/generators/chat/test_hugging_face_api.py b/test/components/generators/chat/test_hugging_face_api.py
index a4db4da2b4..5a8be6c3c5 100644
--- a/test/components/generators/chat/test_hugging_face_api.py
+++ b/test/components/generators/chat/test_hugging_face_api.py
@@ -687,7 +687,6 @@ def test_convert_hfapi_tool_calls_invalid_type_arguments(self):
                     },
                     index=0,
                     start=True,
-                    finish_reason=None,
                 ),
                 [],
             ),
@@ -732,7 +731,6 @@ def test_convert_hfapi_tool_calls_invalid_type_arguments(self):
                         "model": "microsoft/Phi-3.5-mini-instruct",
                         "usage": {"completion_tokens": 2, "prompt_tokens": 21},
                     },
-                    finish_reason=None,
                 ),
                 [0, 1],
             ),
diff --git a/test/components/generators/chat/test_openai.py b/test/components/generators/chat/test_openai.py
index 3fdd53bb34..dfbb2a39d5 100644
--- a/test/components/generators/chat/test_openai.py
+++ b/test/components/generators/chat/test_openai.py
@@ -986,7 +986,6 @@ def streaming_chunks():
                 "received_at": ANY,
                 "usage": None,
             },
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1008,7 +1007,6 @@ def streaming_chunks():
             index=0,
             tool_calls=[ToolCallDelta(tool_name="weather", id="call_zcvlnVaTeJWRjLAFfYxX69z4", index=0)],
             start=True,
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1022,7 +1020,6 @@ def streaming_chunks():
             },
             index=0,
             tool_calls=[ToolCallDelta(arguments='{"ci', index=0)],
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1036,7 +1033,6 @@ def streaming_chunks():
             },
             index=0,
             tool_calls=[ToolCallDelta(arguments='ty": ', index=0)],
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1050,7 +1046,6 @@ def streaming_chunks():
             },
             index=0,
             tool_calls=[ToolCallDelta(arguments='"Paris', index=0)],
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1064,7 +1059,6 @@ def streaming_chunks():
             },
             index=0,
             tool_calls=[ToolCallDelta(arguments='"}', index=0)],
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1086,7 +1080,6 @@ def streaming_chunks():
             index=1,
             tool_calls=[ToolCallDelta(tool_name="weather", id="call_C88m67V16CrETq6jbNXjdZI9", index=1)],
             start=True,
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1100,7 +1093,6 @@ def streaming_chunks():
             },
             index=1,
             tool_calls=[ToolCallDelta(arguments='{"ci', index=1)],
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1114,7 +1106,6 @@ def streaming_chunks():
             },
             index=1,
             tool_calls=[ToolCallDelta(arguments='ty": ', index=1)],
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1128,7 +1119,6 @@ def streaming_chunks():
             },
             index=1,
             tool_calls=[ToolCallDelta(arguments='"Berli', index=1)],
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1142,7 +1132,6 @@ def streaming_chunks():
             },
             index=1,
             tool_calls=[ToolCallDelta(arguments='n"}', index=1)],
-            finish_reason=None,
         ),
         StreamingChunk(
             content="",
@@ -1174,7 +1163,6 @@ def streaming_chunks():
                     "prompt_tokens_details": {"audio_tokens": 0, "cached_tokens": 0},
                 },
             },
-            finish_reason=None,
         ),
     ]