diff --git a/autoarena/judge/anthropic.py b/autoarena/judge/anthropic.py
index fdce1354..e207bdc2 100644
--- a/autoarena/judge/anthropic.py
+++ b/autoarena/judge/anthropic.py
@@ -1,8 +1,7 @@
 from autoarena.api import api
 from autoarena.api.api import JudgeType
 from autoarena.judge.base import AutomatedJudge
-from autoarena.judge.utils import get_user_prompt, rate_limit
-from tests.unit.judge.test_utils import DEFAULT_BATCH_SIZE
+from autoarena.judge.utils import get_user_prompt, rate_limit, DEFAULT_BATCH_SIZE
 
 
 class AnthropicJudge(AutomatedJudge):
diff --git a/autoarena/judge/cohere.py b/autoarena/judge/cohere.py
index 5ca4e4fc..c251e4a4 100644
--- a/autoarena/judge/cohere.py
+++ b/autoarena/judge/cohere.py
@@ -1,8 +1,7 @@
 from autoarena.api import api
 from autoarena.api.api import JudgeType
 from autoarena.judge.base import AutomatedJudge
-from autoarena.judge.utils import get_user_prompt, rate_limit
-from tests.unit.judge.test_utils import DEFAULT_BATCH_SIZE
+from autoarena.judge.utils import get_user_prompt, rate_limit, DEFAULT_BATCH_SIZE
 
 
 class CohereJudge(AutomatedJudge):
diff --git a/autoarena/judge/executor.py b/autoarena/judge/executor.py
index a513ff5c..18606438 100644
--- a/autoarena/judge/executor.py
+++ b/autoarena/judge/executor.py
@@ -8,7 +8,7 @@
 
 from autoarena.api import api
 from autoarena.judge.base import Judge
-from tests.unit.judge.test_utils import DEFAULT_BATCH_SIZE
+from autoarena.judge.utils import DEFAULT_BATCH_SIZE
 
 
 # TODO: this interface is a little gnarly as callers need to deal with responses coming back in any order
diff --git a/autoarena/judge/gemini.py b/autoarena/judge/gemini.py
index d56c2c2c..9fd03e1f 100644
--- a/autoarena/judge/gemini.py
+++ b/autoarena/judge/gemini.py
@@ -1,8 +1,7 @@
 from autoarena.api import api
 from autoarena.api.api import JudgeType
 from autoarena.judge.base import AutomatedJudge
-from autoarena.judge.utils import get_user_prompt, JOINED_PROMPT_TEMPLATE, rate_limit
-from tests.unit.judge.test_utils import DEFAULT_BATCH_SIZE
+from autoarena.judge.utils import get_user_prompt, JOINED_PROMPT_TEMPLATE, rate_limit, DEFAULT_BATCH_SIZE
 
 
 class GeminiJudge(AutomatedJudge):
diff --git a/autoarena/judge/openai.py b/autoarena/judge/openai.py
index edc88846..45ae93a8 100644
--- a/autoarena/judge/openai.py
+++ b/autoarena/judge/openai.py
@@ -1,8 +1,7 @@
 from autoarena.api import api
 from autoarena.api.api import JudgeType
 from autoarena.judge.base import AutomatedJudge
-from autoarena.judge.utils import get_user_prompt, rate_limit
-from tests.unit.judge.test_utils import DEFAULT_BATCH_SIZE
+from autoarena.judge.utils import get_user_prompt, rate_limit, DEFAULT_BATCH_SIZE
 
 
 class OpenAIJudge(AutomatedJudge):
diff --git a/autoarena/judge/utils.py b/autoarena/judge/utils.py
index e1998034..1b5f863c 100644
--- a/autoarena/judge/utils.py
+++ b/autoarena/judge/utils.py
@@ -39,6 +39,9 @@
 
 ACCEPTABLE_RESPONSES = {"A", "B", "-"}
 
+# this is more or less a constant as it is baked into the @rate_limit decorators applied to proprietary judges
+DEFAULT_BATCH_SIZE = 8
+
 
 def get_user_prompt(h2h: api.HeadToHead) -> str:
     return USER_PROMPT_TEMPLATE.format(prompt=h2h.prompt, response_a=h2h.response_a, response_b=h2h.response_b)
diff --git a/tests/unit/judge/test_utils.py b/tests/unit/judge/test_utils.py
index d76f2ff1..c231db52 100644
--- a/tests/unit/judge/test_utils.py
+++ b/tests/unit/judge/test_utils.py
@@ -8,9 +8,6 @@
 from autoarena.judge.base import Judge
 from autoarena.judge.utils import CleaningJudge, RetryingJudge, FixingJudge, ABShufflingJudge, rate_limit
 
-# this is more or less a constant as it is baked into the @rate_limit decorators applied to proprietary judges
-DEFAULT_BATCH_SIZE = 8
-
 
 class DummyJudge(Judge):
     def __init__(self, winners: list[str]):