scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-06 02:48:13 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 103

fix(cost): bill audio input tokens at base rate when model has no audio price

Browse source 
Audio prompt tokens are reported separately from text tokens and subtracted from the text bucket before pricing. When a model ships without input_cost_per_audio_token (37 shipped models including gemini-2.5-pro, gemini-3-pro-preview, and xai grok-4-1-fast), those audio tokens were billed at $0 instead of falling back to input_cost_per_token. Image and video input tokens already fall back to the base rate, and output audio falls back to the base completion rate; input audio was the only modality that silently zeroed.
This commit is contained in:
Anuj7411 2026-08-26 19:29:21 +05:30
parent d2d158f271
commit 3e9bd6db8a
2 changed files with 66 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
litellm/litellm_core_utils/llm_cost_calc/utils.py
View file

@ -662,7 +662,11 @@ def _calculate_input_cost(
### AUDIO COST
if prompt_tokens_details["audio_tokens"]:
audio_cost_key: Final = _get_service_tier_cost_key("input_cost_per_audio_token", service_tier)
tier_audio_cost_key: Final = _get_service_tier_cost_key("input_cost_per_audio_token", service_tier)
has_audio_price: Final = (
model_info.get(tier_audio_cost_key) is not None or model_info.get("input_cost_per_audio_token") is not None
)
audio_cost_key: Final = tier_audio_cost_key if has_audio_price else "input_cost_per_token"
prompt_cost += calculate_cost_component(model_info, audio_cost_key, prompt_tokens_details["audio_tokens"])
### IMAGE TOKEN COST

61
tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py
View file

@ -1449,6 +1449,67 @@ def test_string_cost_values_edge_cases():
assert round(completion_cost, 12) == round(expected_completion_cost, 12)
def test_audio_input_tokens_fall_back_to_base_rate_when_no_audio_price():
"""Audio input tokens must bill at input_cost_per_token when the model has no audio rate.
Regression: models like gemini-2.5-pro report audio prompt tokens separately from text
but ship without input_cost_per_audio_token, which previously billed those tokens at $0.
"""
from unittest.mock import patch
mock_model_info = {
"input_cost_per_token": 1.25e-6,
"output_cost_per_token": 1e-5,
}
usage = Usage(
prompt_tokens=9700,
completion_tokens=0,
total_tokens=9700,
prompt_tokens_details=PromptTokensDetailsWrapper(
audio_tokens=9600, cached_tokens=0, text_tokens=100, image_tokens=None
),
)
with patch(
"litellm.litellm_core_utils.llm_cost_calc.utils.get_model_info",
return_value=mock_model_info,
):
prompt_cost, _ = generic_cost_per_token(model="gemini-2.5-pro", usage=usage, custom_llm_provider="vertex_ai")
expected_prompt_cost = (100 + 9600) * 1.25e-6
assert round(prompt_cost, 12) == round(expected_prompt_cost, 12)
def test_audio_input_tokens_use_audio_rate_when_present():
"""When input_cost_per_audio_token exists it takes precedence over the base rate."""
from unittest.mock import patch
mock_model_info = {
"input_cost_per_token": 1.25e-6,
"input_cost_per_audio_token": 5e-6,
"output_cost_per_token": 1e-5,
}
usage = Usage(
prompt_tokens=9700,
completion_tokens=0,
total_tokens=9700,
prompt_tokens_details=PromptTokensDetailsWrapper(
audio_tokens=9600, cached_tokens=0, text_tokens=100, image_tokens=None
),
)
with patch(
"litellm.litellm_core_utils.llm_cost_calc.utils.get_model_info",
return_value=mock_model_info,
):
prompt_cost, _ = generic_cost_per_token(model="gemini-2.5-pro", usage=usage, custom_llm_provider="vertex_ai")
expected_prompt_cost = 100 * 1.25e-6 + 9600 * 5e-6
assert round(prompt_cost, 12) == round(expected_prompt_cost, 12)
def test_string_cost_values_with_threshold():
"""Test that string cost values work correctly with threshold pricing."""
from unittest.mock import patch