fix(chat): expose stream_batch content for every response

CHANGELOG.md

@@ -1,6 +1,8 @@
 # Changelog
 
 ## [Unreleased]
+### Fixed
+- `stream_batch` now materializes each chunk's content into the shared response proto before yielding, so every response (not just the first) shows accumulated content when read during streaming or in any read order.
 
 ## [v1.19.0](https://github.com/xai-org/xai-sdk-python/releases/tag/v1.19.0) - 2026-08-18
 ### Added

src/xai_sdk/aio/chat.py

@@ -321,6 +321,10 @@ async def stream_batch(self, n: int) -> AsyncIterator[tuple[Sequence[Response],
                     first_chunk_received = True
 
                 responses[0].process_chunk(chunk)
+                # All chunk deltas for every output index accumulate in the first
+                # response's buffers; materialize them into the shared proto so
+                # every response sees current content regardless of read order.
+                responses[0]._sync_buffers_to_proto()
                 yield responses, [Chunk(chunk, i) for i in range(n)]
 
             span.set_attributes(self._make_span_response_attributes(responses))

src/xai_sdk/sync/chat.py

@@ -314,6 +314,10 @@ def stream_batch(self, n: int) -> Iterator[tuple[Sequence[Response], Sequence[Ch
                     first_chunk_received = True
 
                 responses[0].process_chunk(chunk)
+                # All chunk deltas for every output index accumulate in the first
+                # response's buffers; materialize them into the shared proto so
+                # every response sees current content regardless of read order.
+                responses[0]._sync_buffers_to_proto()
                 yield responses, [Chunk(chunk, i) for i in range(n)]
 
             span.set_attributes(self._make_span_response_attributes(responses))

tests/aio/chat_test.py

@@ -136,6 +136,27 @@ async def test_streaming_batch(client):
     assert last_response[1].content == "Hello, this is a test response!"
 
 
+@pytest.mark.asyncio(loop_scope="session")
+async def test_streaming_batch_each_response_populated_during_stream(client):
+    """Reading a later response's content during the stream must show accumulated content.
+
+    Regression: stream_batch accumulated all deltas in the first response's
+    buffers and only materialized them into the shared proto when response[0]
+    was read (or the telemetry span closed). A caller streaming per-option
+    content from response[1] alone got an empty string for the whole stream.
+    """
+    chat = client.chat.create("grok-3-latest")
+    chat.append(user("test message"))
+    stream = chat.stream_batch(2)
+
+    observed = []
+    async for responses, _ in stream:
+        observed.append(responses[1].content)
+
+    assert observed
+    assert observed[-1] == "Hello, this is a test response!"
+
+
 @pytest.mark.asyncio(loop_scope="session")
 async def test_deferred(client):
     chat = client.chat.create("grok-3-latest")

tests/sync/chat_test.py

@@ -126,6 +126,26 @@ def test_streaming_batch(client: Client):
     assert last_response[1].content == "Hello, this is a test response!"
 
 
+def test_streaming_batch_each_response_populated_during_stream(client: Client):
+    """Reading a later response's content during the stream must show accumulated content.
+
+    Regression: stream_batch accumulated all deltas in the first response's
+    buffers and only materialized them into the shared proto when response[0]
+    was read (or the telemetry span closed). A caller streaming per-option
+    content from response[1] alone got an empty string for the whole stream.
+    """
+    chat = client.chat.create("grok-3-latest")
+    chat.append(user("test message"))
+    stream = chat.stream_batch(2)
+
+    observed = []
+    for responses, _ in stream:
+        observed.append(responses[1].content)
+
+    assert observed
+    assert observed[-1] == "Hello, this is a test response!"
+
+
 def test_function_calling(client: Client):
     chat = client.chat.create(
         "grok-3-latest",