diff --git a/llm/utils.py b/llm/utils.py
index a47984d2..e9853185 100644
--- a/llm/utils.py
+++ b/llm/utils.py
@@ -147,9 +147,9 @@ def remove_empty_and_zero(obj):
 def token_usage_string(input_tokens, output_tokens, token_details) -> str:
     bits = []
     if input_tokens is not None:
-        bits.append(f"{input_tokens} input")
+        bits.append(f"{format(input_tokens, ',')} input")
     if output_tokens is not None:
-        bits.append(f"{output_tokens} output")
+        bits.append(f"{format(output_tokens, ',')} output")
     if token_details:
         bits.append(json.dumps(token_details))
     return ", ".join(bits)
diff --git a/tests/test_cli_openai_models.py b/tests/test_cli_openai_models.py
index 8bd45338..3d0a7c16 100644
--- a/tests/test_cli_openai_models.py
+++ b/tests/test_cli_openai_models.py
@@ -174,8 +174,8 @@ def test_gpt4o_mini_sync_and_async(monkeypatch, tmpdir, httpx_mock, async_, usag
                 }
             ],
             "usage": {
-                "prompt_tokens": 10,
-                "completion_tokens": 2,
+                "prompt_tokens": 1000,
+                "completion_tokens": 2000,
                 "total_tokens": 12,
             },
             "system_fingerprint": "fp_49254d0e9b",
@@ -192,7 +192,7 @@ def test_gpt4o_mini_sync_and_async(monkeypatch, tmpdir, httpx_mock, async_, usag
     assert result.exit_code == 0
     assert result.output == "Ho ho ho\n"
     if usage:
-        assert result.stderr == "Token usage: 10 input, 2 output\n"
+        assert result.stderr == "Token usage: 1,000 input, 2,000 output\n"
     # Confirm it was correctly logged
     assert log_db.exists()
     db = sqlite_utils.Database(str(log_db))