fix: suppress misleading register_model unresolved-cost warnings for entries without custom pricing

[devin-ai-integration]

TLDR

Problem this solves:

• Startup spams register_model unresolved-cost warnings for deployments with no custom pricing
• Warnings name opaque sha256 deployment ids instead of the model
• Customers read this as "cost tracking is broken" when it is not

How it solves it:

• Warn only when the registered entry carries explicit custom pricing
• Warning now says only cache cost fields default to 0
• Router and per-request registrations pass the readable model name for the warning

User Flow

Before: a proxy admin with Bedrock models configured (with or without explicit pricing) sees scary unresolved-cost warnings at every boot

1. They start the proxy with a config listing Bedrock and Azure deployments
2. The startup log prints register_model: model=d3582bc5... not in built-in cost map ... cache cost fields will default to 0 for hashed ids and for deployments that have no custom pricing at all
3. They check http://localhost:4000/ui/?page=models and see input/output/cache pricing configured correctly, and requests are billed correctly, so the warning is noise

After: the same boot is quiet unless a deployment genuinely has incomplete custom pricing

1. They start the proxy with the same config
2. No register_model warning is printed for deployments without custom pricing or whose backend model resolves in the built-in cost map
3. If a deployment sets input/output pricing on a model the cost map does not know, one warning is printed naming the readable model (never a hash) and saying only cache cost fields default to 0
4. POST http://localhost:4000/v1/chat/completions returns 200 with the same x-litellm-response-cost as before

Relevant issues

Fixes #32484

Linear ticket

Resolves LIT-6318

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

Shared setup: live proxy on localhost:4000, LITELLM_LOCAL_MODEL_COST_MAP=True, real AWS Bedrock credentials. Main config has a Bedrock deployment with explicit input/output/cache_read pricing plus an inference profile id, an Azure deployment with a base_model, and a plain Bedrock invoke deployment. Edge config has one Bedrock deployment with custom input/output pricing on a made-up model name (bedrock/lit6318-totally-made-up-model) so pricing is genuinely incomplete

Before (86ef1fb)

Boot with configured pricing emits opaque-hash warnings

1. litellm --config lit6318_config.yaml --port 4000, then grep register_model proxy.log
2. Three warnings, two naming sha256 hashes, all claiming the model is "not in built-in cost map ... cache cost fields will default to 0":

register_model: model=d3582bc536fabcb94e681e7dc4dfe2e38dd4b9fa511892c44bffe14f785cdf43 not in built-in cost map and no prefix/region variant matched; cache cost fields will default to 0. To track cache cost, add cache_creation_input_token_cost and cache_read_input_token_cost to model_info
register_model: model=azure/my-deployment-name not in built-in cost map and no prefix/region variant matched; cache cost fields will default to 0. To track cache cost, add cache_creation_input_token_cost and cache_read_input_token_cost to model_info
register_model: model=e42c3315d3a058c7b54bb3d83293c09d2f704847ff0b738bb8f6101ab4d7b310 not in built-in cost map and no prefix/region variant matched; cache cost fields will default to 0. To track cache cost, add cache_creation_input_token_cost and cache_read_input_token_cost to model_info

Live Bedrock request is billed correctly despite the warnings

1. curl -sD - http://localhost:4000/v1/chat/completions -H "Authorization: Bearer sk-1234" -H "Content-Type: application/json" -d '{"model":"bedrock-invoke-haiku","messages":[{"role":"user","content":"Say OK"}],"max_tokens":10}'
2. HTTP/1.1 200 OK with x-litellm-response-cost: 3.19e-05 and the completion "OK", proving the warnings were noise

Genuinely incomplete custom pricing warns with a hash

1. Boot the edge config, then grep register_model proxy.log
2. Two warnings for the one deployment, one naming the opaque hash:

register_model: model=80e6f3fd188fd7605188939d27a76b6b0718158ca48b59acbef5bb46b1a2afcf not in built-in cost map and no prefix/region variant matched; cache cost fields will default to 0. ...
register_model: model=bedrock/lit6318-totally-made-up-model not in built-in cost map and no prefix/region variant matched; cache cost fields will default to 0. ...

After (d321dc6)

Boot with configured pricing emits opaque-hash warnings

1. litellm --config lit6318_config.yaml --port 4000 (byte-identical config), then grep register_model proxy.log
2. Zero register_model warnings

Live Bedrock request is billed correctly despite the warnings

1. Same curl as before
2. HTTP/1.1 200 OK with the identical x-litellm-response-cost: 3.19e-05, so per-request spend is unchanged

Genuinely incomplete custom pricing warns with a hash

1. Boot the same edge config, then grep register_model proxy.log
2. Exactly one precise warning, naming the readable model instead of the hash and scoping the impact to cache fields:

register_model: model=bedrock/lit6318-totally-made-up-model has custom pricing but not in built-in cost map and no prefix/region variant matched; cache_creation_input_token_cost and cache_read_input_token_cost will default to 0 for this model (input/output cost tracking is unaffected). To track cache cost, add them to model_info

Type

Bug Fix

Caveats (if any)

Low

• Per-request re-registration of a custom-priced deployment can still warn once
• That warning now names the readable provider/model key, not the hash

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

Link to Devin session: https://app.devin.ai/sessions/7f184ca33c3f4ac08bcbc818441b74c0
Open in Devin Desktop: https://app.devin.ai/desktop/session/7f184ca33c3f4ac08bcbc818441b74c0?variant=devin
Requested by: @yassin-berriai

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

No action taken on #38542 — repo and author (devin-ai-integration[bot]) match, but the PR has no labels at all, so the required enterprise label is absent. No GitHub or Linear changes made.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

The PR suppresses unresolved-cache-pricing warnings for registrations without explicit custom pricing and replaces opaque deployment IDs with readable model names.

• Limits warnings to unmatched entries with explicit flat input/output pricing and missing cache pricing.
• Excludes tiered pricing, whose cache rates fall back to tier input rates.
• Passes readable model names when router and request paths register opaque deployment IDs.
• Adds regression coverage for silent registrations and warning display names.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
litellm/utils.py	Narrows the warning condition and accurately describes the zero-valued cache pricing fallback for non-tiered custom pricing.
litellm/router.py	Supplies the configured backend model as the warning display name when registering deployment IDs.
litellm/main.py	Supplies the shared provider/model key as the display name during per-request router registration.
tests/test_litellm/test_register_model_custom_pricing.py	Adds focused coverage for silent unpriced and tiered registrations plus readable router warning names.

_{Reviews (2): Last reviewed commit: "fix: do not warn about zero cache costs ..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[devin-ai-integration]

Fixed in 992b74b: tiered entries no longer warn since tiered cache reads fall back to the input rate, and a regression test covers it

[devin-ai-integration]

@greptileai

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_fix_register_model_unresolved_cost_warnings (d321dc6) with litellm_internal_staging (ca9007b)}