diff --git a/backend/api/routers/dub_export.py b/backend/api/routers/dub_export.py
index 63087d46..042dcf6a 100644
--- a/backend/api/routers/dub_export.py
+++ b/backend/api/routers/dub_export.py
@@ -1657,11 +1657,8 @@ async def dub_download_audio(
 
 
 def _format_srt_time(seconds):
-    h = int(seconds // 3600)
-    m = int((seconds % 3600) // 60)
-    s = int(seconds % 60)
-    ms = int((seconds % 1) * 1000)
-    return f"{h:02d}:{m:02d}:{s:02d},{ms:03d}"
+    from services.srt_parser import format_cue_timestamp
+    return format_cue_timestamp(seconds, ",")
 
 def _pick_subtitle_text(seg: dict, dual: bool) -> str:
     """One line per subtitle cue, unless dual=true and an original exists.
@@ -1745,11 +1742,8 @@ async def dub_export_srt(
     )
 
 def _format_vtt_time(seconds):
-    h = int(seconds // 3600)
-    m = int((seconds % 3600) // 60)
-    s = int(seconds % 60)
-    ms = int((seconds % 1) * 1000)
-    return f"{h:02d}:{m:02d}:{s:02d}.{ms:03d}"
+    from services.srt_parser import format_cue_timestamp
+    return format_cue_timestamp(seconds, ".")
 
 @router.get("/dub/vtt/{job_id}")
 @router.get("/dub/vtt/{job_id}/{filename}")
diff --git a/backend/api/routers/openai_compat.py b/backend/api/routers/openai_compat.py
index 03e597fe..7ec030bc 100644
--- a/backend/api/routers/openai_compat.py
+++ b/backend/api/routers/openai_compat.py
@@ -721,17 +721,11 @@ def list_voices():
 
 def _format_ts_srt(seconds: float) -> str:
     """Format seconds as SRT timestamp: HH:MM:SS,mmm"""
-    h = int(seconds // 3600)
-    m = int((seconds % 3600) // 60)
-    s = int(seconds % 60)
-    ms = int((seconds % 1) * 1000)
-    return f"{h:02d}:{m:02d}:{s:02d},{ms:03d}"
+    from services.srt_parser import format_cue_timestamp
+    return format_cue_timestamp(seconds, ",")
 
 
 def _format_ts_vtt(seconds: float) -> str:
     """Format seconds as VTT timestamp: HH:MM:SS.mmm"""
-    h = int(seconds // 3600)
-    m = int((seconds % 3600) // 60)
-    s = int(seconds % 60)
-    ms = int((seconds % 1) * 1000)
-    return f"{h:02d}:{m:02d}:{s:02d}.{ms:03d}"
+    from services.srt_parser import format_cue_timestamp
+    return format_cue_timestamp(seconds, ".")
diff --git a/backend/services/srt_parser.py b/backend/services/srt_parser.py
index 3b26e818..357cb87a 100644
--- a/backend/services/srt_parser.py
+++ b/backend/services/srt_parser.py
@@ -126,3 +126,19 @@ def parse_srt(content: str) -> SrtParseResult:
         for i, seg in enumerate(out)
     ]
     return SrtParseResult(segments=segments, skipped_cues=skipped, dropped_overlaps=dropped)
+
+
+def format_cue_timestamp(seconds: float, ms_separator: str) -> str:
+    """`HH:MM:SS<sep>mmm` for `seconds`, rounded to the millisecond.
+
+    Rounds the whole value once, then splits it, so a time that is not exact
+    in binary (2.3 is 2.29999...) stays 2.300 instead of truncating to 2.299,
+    which moved every such cue a millisecond early on export, and 59.9996
+    carries to the next second instead of printing `,1000`. SRT separates
+    the milliseconds with `,`; WebVTT with `.`.
+    """
+    total_ms = int(round(seconds * 1000))
+    h, rem = divmod(total_ms, 3_600_000)
+    m, rem = divmod(rem, 60_000)
+    s, ms = divmod(rem, 1000)
+    return f"{h:02d}:{m:02d}:{s:02d}{ms_separator}{ms:03d}"
diff --git a/tests/test_subtitle_timestamp_rounding.py b/tests/test_subtitle_timestamp_rounding.py
new file mode 100644
index 00000000..d59fd789
--- /dev/null
+++ b/tests/test_subtitle_timestamp_rounding.py
@@ -0,0 +1,93 @@
+"""Exported subtitle cue times keep their milliseconds.
+
+The SRT/VTT formatters truncated ``(seconds % 1) * 1000``. Most decimal times
+are not exact in binary (2.3 is 2.29999…), so a cue imported as
+``00:00:02,300`` exported as ``00:00:02,299`` — every such cue moved a
+millisecond early in the dub SRT/VTT downloads, the burned-in subtitles and
+the OpenAI-compatible transcription's srt/vtt formats.
+"""
+from __future__ import annotations
+
+import os
+import uuid
+
+import pytest
+
+os.environ.setdefault("OMNIVOICE_MODEL", "test")
+
+# (seconds, SRT form). 2.3 / 4.1 / 70.7 are the binary-inexact values an SRT
+# import produces; 59.9996 must carry into the next second, never ",1000".
+_CASES = [
+    (0.0, "00:00:00,000"),
+    (2.3, "00:00:02,300"),
+    (4.1, "00:00:04,100"),
+    (70.7, "00:01:10,700"),
+    (3661.123, "01:01:01,123"),
+    (59.9996, "00:01:00,000"),
+]
+
+
+@pytest.mark.parametrize("seconds,expected", _CASES)
+def test_dub_srt_and_vtt_times_round_to_the_millisecond(seconds, expected):
+    from api.routers.dub_export import _format_srt_time, _format_vtt_time
+
+    assert _format_srt_time(seconds) == expected
+    assert _format_vtt_time(seconds) == expected.replace(",", ".")
+
+
+@pytest.mark.parametrize("seconds,expected", _CASES)
+def test_openai_compat_srt_and_vtt_times_round_to_the_millisecond(seconds, expected):
+    from api.routers.openai_compat import _format_ts_srt, _format_ts_vtt
+
+    assert _format_ts_srt(seconds) == expected
+    assert _format_ts_vtt(seconds) == expected.replace(",", ".")
+
+
+@pytest.fixture()
+def imported_job():
+    """A dub job carrying the cue times an imported .srt produces."""
+    from services.dub_pipeline import _dub_jobs
+    from services.srt_parser import parse_srt
+
+    segments = parse_srt(
+        "1\n00:00:02,300 --> 00:00:04,100\nFirst\n\n"
+        "2\n00:01:10,700 --> 00:01:12,900\nSecond\n"
+    ).segments
+    job_id = str(uuid.uuid4())[:8]
+    _dub_jobs[job_id] = {
+        "video_path": "/nonexistent/original.mp4",
+        "duration": 80.0,
+        "filename": "imported.mp4",
+        "segments": segments,
+    }
+    yield job_id, _dub_jobs[job_id]
+    _dub_jobs.pop(job_id, None)
+
+
+_SRT_TIMINGS = ["00:00:02,300 --> 00:00:04,100", "00:01:10,700 --> 00:01:12,900"]
+
+
+def test_imported_srt_cue_times_survive_the_srt_and_vtt_downloads(imported_job):
+    from fastapi.testclient import TestClient
+    from main import app
+
+    job_id, _ = imported_job
+    client = TestClient(app, client=("127.0.0.1", 50000))
+
+    srt = client.get(f"/dub/srt/{job_id}")
+    assert srt.status_code == 200
+    assert [line for line in srt.text.splitlines() if "-->" in line] == _SRT_TIMINGS
+
+    vtt = client.get(f"/dub/vtt/{job_id}")
+    assert vtt.status_code == 200
+    assert [line for line in vtt.text.splitlines() if "-->" in line] == [
+        t.replace(",", ".") for t in _SRT_TIMINGS
+    ]
+
+
+def test_imported_srt_cue_times_survive_burn_in(tmp_path, imported_job):
+    from api.routers.dub_export import _write_burn_srt
+
+    _, job = imported_job
+    content = open(_write_burn_srt(job, str(tmp_path), "stamp", dual=False), encoding="utf-8").read()
+    assert [line for line in content.splitlines() if "-->" in line] == _SRT_TIMINGS