diff --git a/litellm/litellm_core_utils/llm_cost_calc/utils.py b/litellm/litellm_core_utils/llm_cost_calc/utils.py
index fdc6127814f..0ba0bd83880 100644
--- a/litellm/litellm_core_utils/llm_cost_calc/utils.py
+++ b/litellm/litellm_core_utils/llm_cost_calc/utils.py
@@ -309,15 +309,14 @@ def _get_token_base_cost(
         float,
         _get_cost_per_unit(model_info, "cache_creation_input_token_cost_above_1hr"),
     )
-    cache_read_cost = cast(float, _get_cost_per_unit(model_info, cache_read_cost_key))
-    # Some model entries only carry the legacy `input_cost_per_token_cache_hit` field
-    # instead of the canonical `cache_read_input_token_cost` (e.g. DeepSeek entries
-    # added before the field was standardized). Without this fallback, cache-hit
-    # tokens silently cost $0 for those models. See BerriAI/litellm#28854.
-    if model_info.get("cache_read_input_token_cost") is None:
-        legacy_cache_hit_cost: Final = model_info.get("input_cost_per_token_cache_hit")
-        if legacy_cache_hit_cost is not None:
-            cache_read_cost = cast(float, _get_cost_per_unit(model_info, "input_cost_per_token_cache_hit"))
+    use_legacy_cache_hit_key: Final = (
+        model_info.get("cache_read_input_token_cost") is None
+        and model_info.get("input_cost_per_token_cache_hit") is not None
+    )
+    effective_cache_read_key: Final = (
+        "input_cost_per_token_cache_hit" if use_legacy_cache_hit_key else cache_read_cost_key
+    )
+    cache_read_cost = cast(float, _get_cost_per_unit(model_info, effective_cache_read_key))
 
     ## CHECK IF ABOVE THRESHOLD
     # Optimization: collect threshold keys first to avoid sorting all model_info keys.