[Bug]: API blocks free (no cost) models when budget is exceeded

[janmett]

What happened?

Problem

We offer both self-hosted on-premises and cloud-based models. Each user is allocated a default budget and has access to all models.

If a user exceeds their budget, the 'ExceededBudget' error is raised, even for our on-premises models which have no configured costs and are completely free to us. The API then becomes unavailable for all models, which is not what I would expect for models that cost nothing.

Expectation

The API should recognise models without costs and allow requests to be made. Cloud models with configured costs (cost > 0) should still be blocked.

Solutions?

Are we missing configuration Options?

If we wanted to fix this issue ourselves, is there a chance it will be merged into litellm? Since a similar issue was marked as "not planned".

Note: The budget in the log output is supposed to provoke this error.

Relevant log output

09:03:41 - LiteLLM Proxy:ERROR: auth_exception_handler.py:79 - litellm.proxy.proxy_server.user_api_key_auth(): Exception occured - ExceededBudget: User=foo over budget. Spend=0.03862624999999999, Budget=0.01
Requester IP Address:172.19.0.1
Traceback (most recent call last):
  File "/usr/lib/python3.13/site-packages/litellm/proxy/auth/user_api_key_auth.py", line 1065, in _user_api_key_auth_builder
    _ = await common_checks(
        ^^^^^^^^^^^^^^^^^^^^
    ...<11 lines>...
    )
    ^
  File "/usr/lib/python3.13/site-packages/litellm/proxy/auth/auth_checks.py", line 149, in common_checks
    raise litellm.BudgetExceededError(
    ...<3 lines>...
    )
litellm.exceptions.BudgetExceededError: ExceededBudget: User=foo over budget. Spend=0.03862624999999999, Budget=0.01
09:03:41 - LiteLLM Proxy:INFO: db_spend_update_writer.py:334 - Writing spend log to db - request_id: 0c65a7cd-8bff-49a3-a8ab-700e412dff36, spend: 0.0
INFO:     172.19.0.1:46236 - "POST /chat/completions HTTP/1.0" 400 Bad Request
09:03:41 - LiteLLM Proxy:INFO: db_spend_update_writer.py:1090 - Logged request status: failure

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

1.75.5

Twitter / LinkedIn details

No response