Merge pull request #41423 from BerriAI/litellm_strip_dated_snapshot_cost_lookup

fix(cost): resolve dated openai/azure snapshots to their undated cost map entry

litellm/utils.py

@@ -5342,6 +5342,13 @@ def _strip_stable_vertex_version(model_name) -> str:
     return re.sub(r"-\d+$", "", model_name)
 
 
+_DATED_SNAPSHOT_SUFFIX: Final = re.compile(r"-\d{4}-\d{2}-\d{2}$")
+
+
+def _strip_dated_snapshot_suffix(model_name: str) -> str:
+    return _DATED_SNAPSHOT_SUFFIX.sub("", model_name)
+
+
 def _get_base_bedrock_model(model_name) -> str:
     """
     Get the base model from the given model name.
@@ -5389,7 +5396,7 @@ def _strip_model_name(model: str, custom_llm_provider: str | None) -> str:
         strip_finetune: Final = _strip_openai_finetune_model_name(model_name=model)
         return strip_finetune
     else:
-        return model
+        return _strip_dated_snapshot_suffix(model_name=model)
 
 
 # Global case-insensitive lookup map for model_cost (built eagerly at module import)

tests/test_litellm/test_cost_calculator.py

@@ -21,7 +21,9 @@ from litellm.types.llms.openai import OpenAIRealtimeStreamList, ResponseAPIUsage
 from litellm.types.rerank import RerankResponse
 from litellm.types.utils import (
     CallTypes,
+    Choices,
     LiteLLMRealtimeStreamLoggingObject,
+    Message,
     ModelInfo,
     ModelResponse,
     PromptTokensDetailsWrapper,
@@ -109,6 +111,28 @@ def test_completion_cost_uses_response_model_for_dynamic_routing(_local_model_co
     assert cost > 0, "Cost should be calculated using response model"
 
 
+def test_completion_cost_strips_dated_azure_snapshot_model(_local_model_cost_map: None) -> None:
+    dated_response = ModelResponse(
+        model="gpt-5.6-luna-2026-07-09",
+        choices=[Choices(index=0, message=Message(role="assistant", content="hi"), finish_reason="stop")],
+        usage=Usage(prompt_tokens=100, completion_tokens=50, total_tokens=150),
+    )
+    dated_response._hidden_params = {"custom_llm_provider": "azure"}
+
+    undated_response = ModelResponse(
+        model="gpt-5.6-luna",
+        choices=[Choices(index=0, message=Message(role="assistant", content="hi"), finish_reason="stop")],
+        usage=Usage(prompt_tokens=100, completion_tokens=50, total_tokens=150),
+    )
+    undated_response._hidden_params = {"custom_llm_provider": "azure"}
+
+    dated_cost = litellm.completion_cost(completion_response=dated_response)
+    undated_cost = litellm.completion_cost(completion_response=undated_response)
+
+    assert dated_cost == undated_cost
+    assert dated_cost > 0
+
+
 def test_cost_calculator_with_response_cost_in_additional_headers():
     class MockResponse(BaseModel):
         _hidden_params = {"additional_headers": {"llm_provider-x-litellm-response-cost": 1000}}

tests/test_litellm/test_utils.py

@@ -183,6 +183,25 @@ def test_get_model_info_strips_openai_finetune_ids_without_a_custom_suffix(local
     assert info["key"] == "ft:gpt-4o-2024-08-06"
 
 
+@pytest.mark.parametrize(
+    ("model", "custom_llm_provider", "expected_key"),
+    [
+        ("gpt-5.6-luna-2026-07-09", "openai", "gpt-5.6-luna"),
+        ("gpt-5.6-luna-2026-07-09", "azure", "azure/gpt-5.6-luna"),
+    ],
+)
+def test_get_model_info_falls_back_from_dated_snapshot_to_undated_entry(
+    local_model_cost_map, model, custom_llm_provider, expected_key
+):
+    info = litellm.get_model_info(model=model, custom_llm_provider=custom_llm_provider)
+    assert info["key"] == expected_key
+
+
+def test_get_model_info_prefers_exact_dated_key_over_stripped(local_model_cost_map):
+    info = litellm.get_model_info(model="gpt-4o-2024-08-06", custom_llm_provider="openai")
+    assert info["key"] == "gpt-4o-2024-08-06"
+
+
 def test_check_provider_match_azure_ai_allows_openai_and_azure():
     """
     Test that azure_ai provider can match openai and azure models.