fix(cost): bill audio input once when both per-token and per-second rates exist

[devin-ai-integration]

TLDR

Problem this solves:

• Vertex gemini-embedding-2 audio input is billed twice after the pricing sync
• Sync PR chore(prices): sync Azure, Azure AI, Gemini, OpenAI, Bedrock, Together AI, Fireworks and Vertex prices: 278 models, 59 new, 30 deprecated #41154 adds input_cost_per_audio_token; catalog also keeps input_cost_per_audio_per_second
• Cost calc charged both, so two Vertex embedding tests fail in chore(prices): sync Azure, Azure AI, Gemini, OpenAI, Bedrock, Together AI, Fireworks and Vertex prices: 278 models, 59 new, 30 deprecated #41154 CI

How it solves it:

• When audio tokens have a per-token rate, skip the per-second audio charge
• Per-second audio still applies for models with no per-token audio rate
• Updates the two embedding tests to the per-token audio price from the Vertex page

User Flow

Before: a user embedding an audio clip on Vertex is overcharged for the audio

1. They send POST https://litellm-domain/v1/embeddings with "model": "vertex_ai/gemini-embedding-2" and a 2 second audio file
2. Vertex reports 64 audio tokens and the response comes back 200
3. https://litellm-domain/ui/?page=logs shows $0.000736 for the request, which is 64 tokens at $6.50/1M plus 2 seconds at $0.00016/s stacked together

After: the same request is billed once

1. They send the same POST https://litellm-domain/v1/embeddings with "model": "vertex_ai/gemini-embedding-2" and the same clip
2. Vertex reports 64 audio tokens and the response comes back 200
3. https://litellm-domain/ui/?page=logs shows $0.000416, which is 64 tokens at $6.50/1M

Relevant issues

Fixes the Vertex AI / Run tests failure on #41154 (generated pricing sync PR). Do not edit that PR; this lands independently

Raw provider evidence, fetched with curl -sL "https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing?x=$RANDOM" and parsed from the HTML table for "Gemini Embedding 2 (Unified Multimodal, Preview)":

Input text   | Online $0.20 / 1M tokens  | Batch $0.10 / 1M tokens
Input image  | Online $0.45 / 1M tokens  | Batch $0.225 / 1M tokens
Input video  | Online $12.00 / 1M tokens | Batch $6.00 / 1M tokens
Input audio  | Online $6.50 / 1M tokens  | Batch $3.25 / 1M tokens

Google now prices audio per token ($6.50/1M = 6.5e-6 per token). The older per-second rate in the catalog (0.00016/s at 32 tokens/s = 5e-6 per token) is not what the page publishes, so the per-token rate wins whenever it exists

Affected release

Linear ticket

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)

Screenshots / Proof of Fix

Cost math only, no provider call needed. Shared script cost_check.py:

import litellm
from litellm.types.utils import Usage, PromptTokensDetailsWrapper
from litellm.litellm_core_utils.llm_cost_calc.utils import generic_cost_per_token
u = Usage(prompt_tokens=64, completion_tokens=0, total_tokens=64,
          prompt_tokens_details=PromptTokensDetailsWrapper(text_tokens=0, audio_tokens=64, audio_length_seconds=2.0))
print(generic_cost_per_token(model="gemini-embedding-2", usage=u, custom_llm_provider="vertex_ai")[0])

Before (b67137b)

1. LITELLM_LOCAL_MODEL_COST_MAP=True uv run python cost_check.py
2. Output: 0.000736 (64 * 6.5e-6 + 2 * 0.00016, audio billed twice)

After (e5391c8)

1. LITELLM_LOCAL_MODEL_COST_MAP=True uv run python cost_check.py
2. Output: 0.000416 (64 * 6.5e-6)

Type

🐛 Bug Fix

Caveats (if any)

Low

• Video is still billed per second; the Vertex page now lists video per token, follow-up for the pricing sync parser

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/10c57e7b5dec41e29684a8c42c5c6aab
Open in Devin Desktop: https://app.devin.ai/desktop/session/10c57e7b5dec41e29684a8c42c5c6aab?variant=devin

[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

[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.}

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_vertex_embedding_audio_double_bill (e5391c8) with main (b67137b)}

[greptile-apps]

Greptile Summary

This PR prevents audio input from being charged by both token count and duration when both rates are configured.

• Detects whether nonzero audio-token usage has an applicable per-token rate.
• Skips duration-based audio billing when token-based billing applies.
• Retains duration billing for models without a per-token audio rate.
• Adds focused regression coverage and updates Vertex Gemini embedding expectations to the catalog’s token rate.

Confidence Score: 5/5

The PR appears safe to merge; the billing precedence is consistent with the configured rates and is covered in both directions.

No actionable failures remain: the rate probe and charge use identical fallback semantics, per-second billing remains available when token pricing is absent, and the modified Vertex tests accurately reflect catalog pricing.

Important Files Changed

Filename	Overview
litellm/litellm_core_utils/llm_cost_calc/utils.py	Adds consistent per-token precedence to prevent duplicate audio input charges while preserving per-second fallback.
tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py	Covers both dual-rate token precedence and duration-only fallback behavior.
tests/test_litellm/llms/vertex_ai/gemini_embeddings/test_batch_embed_content_transformation.py	Updates Vertex Gemini embedding cost assertions to the configured per-audio-token rate without weakening modality-accounting checks.

_{Reviews (1): Last reviewed commit: "fix(cost): bill audio input once when bo..." | 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]

Superseded: main moved gemini-embedding-2 to per-token rates (ce83fac, a28ea22, 4a8ec7b) and fixed the same tests. Closing.