fix(router): don't log 'Could not identify azure model' when the deployment name resolves from the cost map

[mihidumh]

Relevant issues

Fixes #33172

Linear ticket

Pre-Submission checklist

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

• I have added meaningful tests
• My PR passes all CI/CD checks (e.g., lint, format, unit tests)
• 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)

Screenshots / Proof of Fix

Production observation (litellm proxy 1.92.0 on Azure, real traffic): every request to a multi-deployment azure group without base_model logs

Could not identify azure model 'gpt-5.4'. Set azure 'base_model' for accurate max tokens, cost tracking, etc.

— thousands of ERROR lines/day — even though azure/gpt-5.4 is an exact key in the shipped cost map and get_router_model_info's existing if model is None: model = _model fallback resolves it correctly. We verified on the live proxy that hand-setting model_info.base_model = litellm_params.model on every such row (56 rows on one env) changes nothing about resolution and only silences the log — i.e. the ERROR was spurious for exactly the deployments that need no operator action.

While investigating I found membership in litellm.model_cost can't be the discriminator on its own: Router.__init__ auto-registers every deployment name into the map as a zeroed stub (register_model: model=... not in built-in cost map ... will default to 0). The check therefore requires the fallback entry to carry usable limits/costs.

Unit tests (new TestAzureBaseModelFallbackLogging in tests/test_litellm/test_router.py):

Before (base 10d5804b3e): test_map_known_deployment_name_resolves_without_error_log FAILS (spurious ERROR fires); other 2 pass.

After (this PR, 261f9546c5):

$ pytest tests/test_litellm/test_router.py::TestAzureBaseModelFallbackLogging -q
3 passed

covering: (1) map-known deployment name → no ERROR + model_info carries the map's max tokens/costs, (2) genuinely unmappable name → ERROR still logged, zeroed-stub return unchanged, (3) explicit base_model still wins.

ruff format --check / ruff check pass on litellm/router.py.

Type

🐛 Bug Fix

Changes

• In get_router_model_info's azure branch: when base_model is unset, check whether the azure/<deployment model name> entry in the cost map carries usable limits/costs (guarding against Router-init's zeroed auto-registration stubs). If so, log at debug — the existing fallback resolution handles it. Otherwise keep the existing error.
• No behavior change to resolution itself — only the logging level is corrected for the already-working path.

[greptile-apps]

Greptile Summary

This PR fixes spurious per-request ERROR log lines for Azure deployments whose model name is a direct key in the built-in cost map (e.g., azure/gpt-4o). The fix is logging-only — no change to resolution logic or returned values.

• In get_router_model_info, when base_model is unset for an Azure deployment, the PR checks whether the azure/<model> key in litellm.model_cost carries a positive max_input_tokens, max_tokens, or input_cost_per_token. If so, it downgrades the log from ERROR to DEBUG; otherwise the ERROR is preserved as before.
• Three unit tests cover the new branching: known deployment → no error log, unmappable name → error preserved, explicit base_model → continues to win.

Confidence Score: 5/5

Safe to merge — the change is restricted to log level selection and does not affect model resolution, cost tracking, or token-limit enforcement.

The change touches only one branch inside get_router_model_info and only adjusts whether a message is logged at ERROR vs DEBUG. The (value or 0) > 0 guards are consistent across all three checked fields, correctly rejecting zeroed stubs and explicit-zero entries. Actual resolution is unchanged. Tests are well-constructed mocks that reproduce the exact before/after behavior described in the PR.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/router.py	Logging-only change in get_router_model_info: downgrades Azure "Could not identify model" ERROR to DEBUG when the deployment name resolves to a cost-map entry with usable (>0) limits or costs. Resolution behavior is unchanged.
tests/test_litellm/test_router.py	Adds TestAzureBaseModelFallbackLogging with three new mock-only tests covering the happy path (no error log), the genuine-miss path (error preserved), and the explicit-base-model path. No real network calls; fixture correctly isolates the cost map.

_{Reviews (2): Last reviewed commit: "fix(router): use consistent positive che..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[mihidumh]

@greptileai — made the _fallback_resolves checks consistent: all three fields (max_input_tokens, max_tokens, input_cost_per_token) now require a positive value, so an entry with explicit-zero limits can no longer suppress the error log.

[mihidumh]

Rebased onto litellm_internal_staging (per the current CONTRIBUTING.md) and retargeted — this had been sitting on litellm_oss_daily_2026_07_20, which was retired in 8a0bb4c two days after this PR was moved onto it, so the CI failures were all base age rather than anything in the diff. Force-pushed because the rebase needed a conflict resolution in tests/test_litellm/test_router.py, where #34564 added tests in the same place; both test blocks are kept, the litellm/router.py change is unchanged.

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing mihidumh:fix/azure-base-model-inference (fa2b896) with litellm_internal_staging (47a7e17)¹}

Footnotes

1. No successful run was found on litellm_internal_staging (5290150) during the generation of this report, so 47a7e17 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[mihidumh]

Closing — superseded by #37869, which merged this branch's commit (authorship preserved) into litellm_internal_staging on 2026-08-21. Thanks @tin-berri.