diff --git a/evalscope/benchmarks/ocr_bench/ocr_bench_v2/spotting_metric.py b/evalscope/benchmarks/ocr_bench/ocr_bench_v2/spotting_metric.py
index c68b5c812..27a2e8b4a 100644
--- a/evalscope/benchmarks/ocr_bench/ocr_bench_v2/spotting_metric.py
+++ b/evalscope/benchmarks/ocr_bench/ocr_bench_v2/spotting_metric.py
@@ -159,12 +159,12 @@ def spotting_evaluation(prediction_list, img_metas):
     if len(res_submit_list) == 0 or len(res_gt_list) == 0:
         return 0
 
-    with open(os.path.join(submit_path, 'res_img_0.txt'), 'w') as f:
+    with open(os.path.join(submit_path, 'res_img_0.txt'), 'w', encoding='utf-8') as f:
         for item in res_submit_list[:-1]:
             f.write(item + '\n')
         f.write(res_submit_list[-1])
 
-    with open(os.path.join(gt_path, 'gt_img_0.txt'), 'w') as f:
+    with open(os.path.join(gt_path, 'gt_img_0.txt'), 'w', encoding='utf-8') as f:
         for item in res_gt_list[:-1]:
             f.write(item + '\n')
         f.write(res_gt_list[-1])
diff --git a/tests/benchmark/test_ocr_bench_v2_spotting.py b/tests/benchmark/test_ocr_bench_v2_spotting.py
new file mode 100644
index 000000000..764f6d58c
--- /dev/null
+++ b/tests/benchmark/test_ocr_bench_v2_spotting.py
@@ -0,0 +1,36 @@
+import builtins
+import zipfile
+
+from evalscope.benchmarks.ocr_bench.ocr_bench_v2 import spotting_metric
+
+
+def test_spotting_files_are_written_as_utf8(tmp_path, monkeypatch):
+    """Regression: the spotting scorer decodes the gt/submission files as UTF-8, so they must be written as UTF-8.
+
+    With the locale default encoding (cp1252, cp1254, ... on Windows) a recognised text such as
+    'Café → Exit' raised UnicodeEncodeError before scoring started.
+    """
+
+    def ansi_open(file, mode='r', *args, **kwargs):
+        if 'b' not in mode:
+            kwargs.setdefault('encoding', 'cp1252')
+        return builtins.open(file, mode, *args, **kwargs)
+
+    received = {}
+
+    def fake_main_evaluation(params, *args, **kwargs):
+        for key in ('g', 's'):
+            with zipfile.ZipFile(params[key]) as zf:
+                received[key] = spotting_metric.rrc_evaluation_funcs.decode_utf8(zf.read(zf.namelist()[0]))
+        return {'method': {'hmean': 1.0}}
+
+    monkeypatch.setattr(spotting_metric, 'open', ansi_open, raising=False)
+    monkeypatch.setattr(spotting_metric, 'DEFAULT_EVALSCOPE_CACHE_DIR', str(tmp_path))
+    monkeypatch.setattr(spotting_metric.rrc_evaluation_funcs, 'main_evaluation', fake_main_evaluation)
+
+    text = 'Café → Exit'
+    doc = {'bbox_list': [[10, 10, 100, 10, 100, 40, 10, 40]], 'content': [text]}
+
+    assert spotting_metric.spotting_evaluation([[10, 10, 100, 40, text]], doc) == 1.0
+    assert received['g'].endswith(text)
+    assert received['s'].endswith(text)