diff --git a/.sampo/changesets/gemini-thinking-blocks.md b/.sampo/changesets/gemini-thinking-blocks.md
new file mode 100644
index 00000000..1ab1aff3
--- /dev/null
+++ b/.sampo/changesets/gemini-thinking-blocks.md
@@ -0,0 +1,5 @@
+---
+pypi/posthog: minor
+---
+
+Capture Gemini thought summaries as `thinking` content blocks. When a request enables `thinking_config.include_thoughts`, parts marked `thought=True` in responses, inputs, and streaming chunks are now formatted as `{"type": "thinking", "thinking": ...}` (matching the Anthropic thinking-block shape PostHog renders as reasoning) instead of plain text blocks.
diff --git a/posthog/ai/gemini/gemini_converter.py b/posthog/ai/gemini/gemini_converter.py
index 9ad711d7..bd82a161 100644
--- a/posthog/ai/gemini/gemini_converter.py
+++ b/posthog/ai/gemini/gemini_converter.py
@@ -47,6 +47,10 @@ def _format_part(part: Any) -> Optional[FormattedContentItem]:
     if not isinstance(plain, dict):
         return {"type": "unknown", "part": str(plain)}
     plain = normalize_part_keys(plain)
+    if plain.get("thought") is True and isinstance(plain.get("text"), str):
+        # Thought summaries (thinking_config.include_thoughts) become thinking
+        # blocks, matching the Anthropic shape the PostHog UI renders as reasoning.
+        return {"type": "thinking", "thinking": plain["text"]}
     if "text" in plain:
         return {"type": "text", "text": plain["text"]}
     if "inline_data" in plain:
@@ -245,7 +249,12 @@ def format_gemini_response(response: Any) -> List[FormattedMessage]:
                         # the _format_part delegation below.
                         text = getattr(part, "text", None)
                         if isinstance(text, str) and text:
-                            content.append({"type": "text", "text": text})
+                            # `is True` so loosely-specced mocks (whose .thought is a
+                            # truthy Mock) still land on the text branch below.
+                            if getattr(part, "thought", None) is True:
+                                content.append({"type": "thinking", "thinking": text})
+                            else:
+                                content.append({"type": "text", "text": text})
                             continue
 
                         if hasattr(part, "function_call") and part.function_call:
diff --git a/posthog/test/ai/gemini/test_gemini_converter.py b/posthog/test/ai/gemini/test_gemini_converter.py
index 3991a8ef..3d03e2fe 100644
--- a/posthog/test/ai/gemini/test_gemini_converter.py
+++ b/posthog/test/ai/gemini/test_gemini_converter.py
@@ -244,3 +244,65 @@ def test_streaming_chunk_with_mixed_text_and_image_captures_both(self):
         assert types_seen == ["text", "image"]
         assert content[0] == {"type": "text", "text": "here: "}
         assert content[1]["inline_data"]["data"] == base64.b64encode(PNG).decode()
+
+
+class TestGeminiThoughtParts:
+    def test_response_thought_part_becomes_thinking_block(self):
+        resp = types.GenerateContentResponse(
+            candidates=[
+                types.Candidate(
+                    content=types.Content(
+                        role="model",
+                        parts=[
+                            types.Part(text="Weighing the options...", thought=True),
+                            types.Part(text='{"verdict": "yes"}'),
+                        ],
+                    ),
+                )
+            ]
+        )
+        out = format_gemini_response(resp)
+        assert out[0]["content"] == [
+            {"type": "thinking", "thinking": "Weighing the options..."},
+            {"type": "text", "text": '{"verdict": "yes"}'},
+        ]
+
+    def test_input_thought_part_becomes_thinking_block(self):
+        # Multi-turn callers append the model's prior content (thoughts included)
+        # back into the conversation, so thought parts also arrive as input.
+        contents = [
+            types.Content(
+                role="model",
+                parts=[
+                    types.Part(text="Considering...", thought=True),
+                    types.Part(text="answer"),
+                ],
+            )
+        ]
+        out = format_gemini_input(contents)
+        assert out[0]["content"] == [
+            {"type": "thinking", "thinking": "Considering..."},
+            {"type": "text", "text": "answer"},
+        ]
+
+    def test_streaming_chunk_thought_part_becomes_thinking_block(self):
+        chunk = types.GenerateContentResponse(
+            candidates=[
+                types.Candidate(
+                    content=types.Content(
+                        role="model",
+                        parts=[types.Part(text="Hmm, ", thought=True)],
+                    )
+                )
+            ]
+        )
+        blocks = extract_gemini_content_from_chunk(chunk)
+        assert blocks == [{"type": "thinking", "thinking": "Hmm, "}]
+
+        out = format_gemini_streaming_output(
+            blocks + [{"type": "text", "text": "done"}]
+        )
+        assert out[0]["content"] == [
+            {"type": "thinking", "thinking": "Hmm, "},
+            {"type": "text", "text": "done"},
+        ]