fix(pricing): add undated azure aliases for gpt-audio-mini and gpt-realtime-mini

[mihidumh]

Relevant issues

Fixes #33170

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 Azure gpt-audio-mini deployment, real audio-modality completions): with no undated azure/gpt-audio-mini map key, spend logs and the x-litellm-response-cost header billed text tokens at $0 while audio tokens billed correctly. We verified by putting the correct prices in litellm_params custom pricing on the deployment row — after which the cost header matched the hand-computed text+audio math exactly ($0.60/M in, $2.40/M out — the same numbers as the dated azure/gpt-audio-mini-2025-10-06 entry this PR mirrors).

Map state before (base 10d5804b3e):

$ jq 'has("azure/gpt-audio-mini"), has("azure/gpt-realtime-mini")' model_prices_and_context_window.json
false
false
$ jq -r 'keys[] | select(test("^azure/gpt-(audio|realtime)-mini"))' model_prices_and_context_window.json
azure/gpt-audio-mini-2025-10-06
azure/gpt-realtime-mini-2025-10-06

After (2f906a3415):

$ pytest tests/test_litellm/test_azure_audio_price_aliases.py -q
2 passed
# on base: 2 failed (ValueError: This model isn't mapped yet)

The new test resolves both undated names through litellm.get_model_info (local cost map fixture, same pattern as test_cloudflare_workers_ai_model_metadata.py) and asserts every cost field matches the dated entry and is non-zero.

Note on the second commit: running the repo's own ci_cd/check_files_match.py sync tool surfaced pre-existing drift on the daily branch (9 entries in the root map missing from the packaged backup — gemini-3-pro-image, zai/glm-5.1, etc.). That sync is kept as a separate chore: commit so the alias fix itself stays a clean 63-line-per-file addition. Happy to drop that commit if you'd rather take the drift separately.

Type

🐛 Bug Fix

Changes

• Add azure/gpt-audio-mini and azure/gpt-realtime-mini to model_prices_and_context_window.json (and the packaged backup), each an exact mirror of its -2025-10-06 dated entry — same relationship as the undated openai gpt-audio-mini / gpt-realtime-mini entries.
• Parametrized test asserting the undated aliases resolve and match the dated entries with non-zero costs.

[greptile-apps]

Greptile Summary

This PR fixes a billing bug (issue #33170) where Azure deployments using the undated gpt-audio-mini or gpt-realtime-mini model names had text tokens billed at $0, because those keys were absent from the model cost map. It adds the two missing aliases as exact mirrors of their -2025-10-06 dated entries and syncs pre-existing drift between the root JSON and the packaged backup.

• Adds azure/gpt-audio-mini and azure/gpt-realtime-mini to both model_prices_and_context_window.json and its backup, each field-for-field identical to the corresponding dated entry — verified by direct inspection of both files.
• Adds two parametrized test functions: one checking the four core COST_FIELDS are equal and non-zero, and a second (test_undated_azure_audio_alias_is_exact_mirror) doing full dictionary equality against litellm.model_cost, covering every field including realtime-specific cache_creation_input_audio_token_cost, cache_read_input_token_cost, and input_cost_per_image.

Confidence Score: 5/5

Safe to merge — the change is additive data only (two new JSON keys) with no modifications to existing entries or runtime logic, and the new test suite guards the critical invariant going forward.

Both undated alias entries are confirmed to be byte-for-byte copies of their dated counterparts in both the root and backup JSON files. The test file is mock-only (no network calls), uses the local cost-map fixture correctly, and the new full-equality test closes the gap identified in the prior review round by covering every field including realtime-specific cache and audio cost keys.

No files require special attention. The backup drift-sync entries are straightforward catch-up additions that were already present in the root map.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Adds azure/gpt-audio-mini and azure/gpt-realtime-mini as exact mirrors of their -2025-10-06 dated counterparts; no other changes to the root map.
litellm/model_prices_and_context_window_backup.json	Adds the two azure aliases (identical to root), then syncs nine previously missing entries (Gemini image models, vertex_ai aliases, openrouter/zai models, token-limit corrections, snowflake adaptive-thinking flag) from the root map into the backup. Both undated azure entries verified to be field-for-field identical to their dated siblings.
tests/test_litellm/test_azure_audio_price_aliases.py	New mock-only test file (uses local cost-map fixture, no network calls). Provides two parametrized tests: COST_FIELDS equality + non-zero guards, and a full dictionary equality check (test_undated_azure_audio_alias_is_exact_mirror) covering every field including realtime-specific cache/audio cost keys.

_{Reviews (2): Last reviewed commit: "test(pricing): assert undated azure audi..." | Re-trigger Greptile}

[mihidumh]

@greptileai — added a full-entry equality test asserting each undated alias exactly mirrors its dated sibling, covering the realtime-specific cost keys COST_FIELDS missed.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing mihidumh:fix/azure-audio-mini-price-aliases (e189a58) 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. ↩

[tin-berri]

Thanks for the fix @mihidumh! I copied this to a litellm_-prefixed branch so the full CI pipeline runs: #37867 (your commits and authorship preserved, rebased onto latest litellm_internal_staging). All checks are green there with a Greptile 5/5 — a maintainer can review/merge from that PR.

[mateo-berri]

Superseded by #37867, which landed your commits with added tests for the root cost map. Thanks for the fix and the report