diff --git a/backend/tests/test_utils_llm_text.py b/backend/tests/test_utils_llm_text.py
new file mode 100644
index 000000000..8a0d72ada
--- /dev/null
+++ b/backend/tests/test_utils_llm_text.py
@@ -0,0 +1,157 @@
+"""Tests for ``deerflow.utils.llm_text``."""
+
+from __future__ import annotations
+
+from deerflow.utils.llm_text import (
+    extract_response_text,
+    strip_markdown_code_fence,
+    strip_think_blocks,
+)
+
+# ---------------------------------------------------------------------------
+# strip_think_blocks
+# ---------------------------------------------------------------------------
+
+
+def test_strip_think_blocks_removes_complete_block() -> None:
+    assert strip_think_blocks("before<think>reasoning</think>after") == "beforeafter"
+
+
+def test_strip_think_blocks_removes_multiline_block() -> None:
+    # DOTALL: the block spans newlines and surrounding whitespace is stripped.
+    text = "answer\n<think>\nmulti\nline\n</think>\n"
+    assert strip_think_blocks(text) == "answer"
+
+
+def test_strip_think_blocks_is_case_insensitive_and_tolerates_close_spacing() -> None:
+    assert strip_think_blocks("<THINK>x</THINK >done") == "done"
+
+
+def test_strip_think_blocks_handles_open_tag_attributes() -> None:
+    assert strip_think_blocks('<think class="a">secret</think>visible') == "visible"
+
+
+def test_strip_think_blocks_removes_multiple_blocks_non_greedy() -> None:
+    # Non-greedy matching removes each block independently, not everything
+    # between the first open and the last close.
+    assert strip_think_blocks("a<think>1</think>b<think>2</think>c") == "abc"
+
+
+def test_strip_think_blocks_response_that_is_only_reasoning_becomes_empty() -> None:
+    assert strip_think_blocks("<think>only</think>") == ""
+
+
+def test_strip_think_blocks_passes_plain_text_through() -> None:
+    assert strip_think_blocks("just text") == "just text"
+
+
+def test_strip_think_blocks_truncates_unclosed_by_default() -> None:
+    # A dangling open tag means the model was truncated mid-thought; drop the
+    # rest of the text so downstream JSON parsers do not choke on it.
+    assert strip_think_blocks("visible answer <think>partial reasoning") == "visible answer"
+
+
+def test_strip_think_blocks_keeps_unclosed_tag_when_truncation_disabled() -> None:
+    text = "visible answer <think>partial reasoning"
+    assert strip_think_blocks(text, truncate_unclosed=False) == text
+
+
+def test_strip_think_blocks_removes_complete_then_truncates_dangling() -> None:
+    assert strip_think_blocks("<think>done</think>keep<think>trunc") == "keep"
+
+
+def test_strip_think_blocks_removes_complete_and_keeps_dangling_when_disabled() -> None:
+    result = strip_think_blocks("<think>done</think>keep<think>trunc", truncate_unclosed=False)
+    assert result == "keep<think>trunc"
+
+
+# ---------------------------------------------------------------------------
+# strip_markdown_code_fence
+# ---------------------------------------------------------------------------
+
+
+def test_strip_markdown_code_fence_unwraps_language_fence() -> None:
+    assert strip_markdown_code_fence('```json\n{"a": 1}\n```') == '{"a": 1}'
+
+
+def test_strip_markdown_code_fence_unwraps_bare_fence() -> None:
+    assert strip_markdown_code_fence("```\nhello\n```") == "hello"
+
+
+def test_strip_markdown_code_fence_ignores_surrounding_whitespace() -> None:
+    assert strip_markdown_code_fence("  ```json\n{}\n```  ") == "{}"
+
+
+def test_strip_markdown_code_fence_preserves_multiline_body() -> None:
+    fenced = "```python\ndef f():\n    return 1\n```"
+    assert strip_markdown_code_fence(fenced) == "def f():\n    return 1"
+
+
+def test_strip_markdown_code_fence_returns_plain_text_unchanged() -> None:
+    assert strip_markdown_code_fence("plain text") == "plain text"
+
+
+def test_strip_markdown_code_fence_ignores_inline_backticks() -> None:
+    assert strip_markdown_code_fence("see `code` here") == "see `code` here"
+
+
+def test_strip_markdown_code_fence_leaves_lone_fence_line_unchanged() -> None:
+    # Fewer than three lines cannot be an opening + body + closing fence.
+    assert strip_markdown_code_fence("```json") == "```json"
+
+
+def test_strip_markdown_code_fence_leaves_unterminated_fence_unchanged() -> None:
+    assert strip_markdown_code_fence("```\ncontent") == "```\ncontent"
+
+
+# ---------------------------------------------------------------------------
+# extract_response_text
+# ---------------------------------------------------------------------------
+
+
+def test_extract_response_text_passes_string_through_verbatim() -> None:
+    # No stripping: the raw string content is returned unchanged.
+    assert extract_response_text("  hi  ") == "  hi  "
+
+
+def test_extract_response_text_joins_string_blocks() -> None:
+    assert extract_response_text(["a", "b"]) == "a\nb"
+
+
+def test_extract_response_text_reads_text_and_output_text_blocks() -> None:
+    content = [
+        {"type": "text", "text": "x"},
+        {"type": "output_text", "text": "y"},
+    ]
+    assert extract_response_text(content) == "x\ny"
+
+
+def test_extract_response_text_ignores_non_text_blocks() -> None:
+    content = [
+        {"type": "tool_use", "text": "ignored"},
+        {"type": "text", "text": "kept"},
+    ]
+    assert extract_response_text(content) == "kept"
+
+
+def test_extract_response_text_mixes_string_and_dict_blocks() -> None:
+    content = ["intro", {"type": "text", "text": "body"}]
+    assert extract_response_text(content) == "intro\nbody"
+
+
+def test_extract_response_text_skips_blocks_with_non_string_text() -> None:
+    content = [{"type": "text", "text": 123}, {"type": "text", "text": "ok"}]
+    assert extract_response_text(content) == "ok"
+
+
+def test_extract_response_text_returns_empty_for_empty_list() -> None:
+    assert extract_response_text([]) == ""
+
+
+def test_extract_response_text_returns_empty_for_none() -> None:
+    assert extract_response_text(None) == ""
+
+
+def test_extract_response_text_stringifies_other_types() -> None:
+    assert extract_response_text(123) == "123"
+    assert extract_response_text({"a": 1}) == str({"a": 1})