From af44f48623abe1e70bcd828ae8efdcdb1fcdac79 Mon Sep 17 00:00:00 2001
From: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Wed, 8 Jul 2026 16:11:08 +0000
Subject: [PATCH] fix(cost): resolve cache pricing via base_model to silence
 spurious register_model warning

---
 litellm/utils.py                              | 21 +++++++++++
 .../test_register_model_custom_pricing.py     | 35 +++++++++++++++++++
 2 files changed, 56 insertions(+)

diff --git a/litellm/utils.py b/litellm/utils.py
index 19c2fe16085..5c80a092402 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -2651,6 +2651,23 @@ def _resolve_builtin_model_cost_entry(key: str, provider: str) -> Optional[Dict[
     return None
 
 
+def _resolve_base_model_cost_entry(base_model: Optional[str], provider: str) -> Optional[Dict[str, Any]]:
+    """Resolve a built-in ``model_cost`` entry from a deployment's ``base_model``.
+
+    Deployments that price via ``model_info.base_model`` (e.g. an Azure
+    deployment aliasing ``azure/text-embedding-3-large``) carry no cost keys on
+    their own key, so ``register_model`` would otherwise warn about missing cache
+    pricing even though the base model fully describes it. Falls back to the
+    prefix/region resolver for base models whose exact key is not present.
+    """
+    if not isinstance(base_model, str) or not base_model:
+        return None
+    entry = litellm.model_cost.get(base_model)
+    if entry is not None and entry.get("litellm_provider") is not None:
+        return dict(entry)
+    return _resolve_builtin_model_cost_entry(key=base_model, provider=provider)
+
+
 def register_model(model_cost: Union[str, dict]):
     """
     Register new / Override existing models (and their pricing) to specific providers.
@@ -2698,6 +2715,10 @@ def register_model(model_cost: Union[str, dict]):
                 existing_model = {}
                 model_cost_key = key
                 builtin_entry = _resolve_builtin_model_cost_entry(key=_key_str, provider=provider)
+                if builtin_entry is None:
+                    builtin_entry = _resolve_base_model_cost_entry(
+                        base_model=value.get("base_model"), provider=provider
+                    )
                 if builtin_entry is not None:
                     for field in _CACHE_PRICING_FIELDS:
                         if value.get(field) is None and builtin_entry.get(field) is not None:
diff --git a/tests/test_litellm/test_register_model_custom_pricing.py b/tests/test_litellm/test_register_model_custom_pricing.py
index ba82bfaadc6..508a89e742c 100644
--- a/tests/test_litellm/test_register_model_custom_pricing.py
+++ b/tests/test_litellm/test_register_model_custom_pricing.py
@@ -439,6 +439,41 @@ def test_register_model_warns_when_no_builtin_match_for_cache_pricing(caplog):
         litellm.model_cost.pop(registered_key, None)
 
 
+def test_register_model_no_warning_when_base_model_resolves_cache_pricing(caplog):
+    """Regression for https://github.com/BerriAI/litellm/issues/32484.
+
+    A deployment priced via ``model_info.base_model`` (e.g. an Azure deployment
+    aliasing ``azure/text-embedding-3-large``) carries no cost keys on its own
+    key, so ``register_model`` used to warn that cache cost fields default to 0
+    even though the base model fully describes pricing. When ``base_model``
+    resolves to a built-in entry the warning must not fire.
+    """
+    import logging
+
+    from litellm._logging import verbose_logger
+
+    registered_key = "azure/my-deployment-name-32484"
+    litellm.model_cost.pop(registered_key, None)
+
+    try:
+        with caplog.at_level(logging.WARNING, logger=verbose_logger.name):
+            litellm.register_model(
+                {
+                    registered_key: {
+                        "litellm_provider": "azure",
+                        "base_model": "azure/text-embedding-3-large",
+                    }
+                }
+            )
+
+        assert not any(
+            registered_key in record.message and "cache_creation_input_token_cost" in record.message
+            for record in caplog.records
+        ), "did not expect a cache-pricing warning when base_model resolves to a built-in entry"
+    finally:
+        litellm.model_cost.pop(registered_key, None)
+
+
 def test_register_model_router_add_deployment_custom_pricing_applies():
     """End-to-end regression for https://github.com/BerriAI/litellm/issues/28336.