fix: increase health check max_tokens from 1 to 16 (#23836)

[hannahmadison]

Relevant issues

Fixes #23836

Pre-Submission checklist

• I have added testing in the tests/test_litellm/ directory
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

Type

🐛 Bug Fix

Changes

Problem

Health checks set max_tokens=1 for non-wildcard models, which causes failures with newer models like GPT-5 that have minimum token requirements. A single token is insufficient for these models to generate a valid response.

Solution

Increased the default max_tokens from 1 to 16 for health check requests. This provides enough tokens for models to generate a meaningful response while keeping the health check lightweight.

Also tightened the wildcard test assertion from a weak disjunctive check (not in or != 1) to strict key-absence (not in), preventing false passes if a future bug sets max_tokens to an arbitrary value for wildcard models.

Changes Made

• litellm/proxy/health_check.py: Changed max_tokens default from 1 to 16
• tests/test_litellm/proxy/test_health_check_max_tokens.py: Updated default assertion to expect 16, tightened wildcard assertion to strict key-absence check

Supersedes #24893 (closed due to branch contamination from upstream merge).

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Greptile Summary

This PR increases the health check max_tokens default from 1 to 16 in litellm/proxy/health_check.py to fix compatibility with newer models (e.g. GPT-5) that require a minimum token budget to return a valid response. Tests are updated to reflect the new default, and the wildcard model assertion is tightened to strict key-absence.

Confidence Score: 5/5

Safe to merge — minimal, well-tested change with no functional regressions.

The change is a single-line bump of a default constant from 1 to 16. Tests are properly updated, the wildcard assertion is now stricter (not weaker), and all tests remain mock-only. No security, data, or backwards-compatibility concerns.

No files require special attention.

Important Files Changed

Filename	Overview
litellm/proxy/health_check.py	Single-line change: max_tokens default for non-wildcard health checks raised from 1 to 16. Logic and branching are unchanged.
tests/test_litellm/proxy/test_health_check_max_tokens.py	Assertions updated to match new default (16) and wildcard test tightened to strict key-absence; all tests use mocks with no real network calls.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[_update_litellm_params_for_health_check] --> B{health_check_max_tokens\nin model_info?}
    B -- Yes --> C[max_tokens = health_check_max_tokens]
    B -- No --> D{"'*' in model or\nhealth_check_model?"}
    D -- Yes wildcard --> E[max_tokens not set]
    D -- No non-wildcard --> F[max_tokens = 16\npreviously 1]
    C --> G[proceed with health check call]
    E --> G
    F --> G

Loading

_{Reviews (3): Last reviewed commit: "fix: increase health check max_tokens fr..." | Re-trigger Greptile}

[hannahmadison]

Hey @Sameerlite Quick check: was #26217 closed intentionally, or as a side effect of litellm_oss_branch being deleted? If the latter, happy to re-open against main. The change itself is a one-line fix for #23836 and was Greptile-approved at 5/5.

[Sameerlite]

It is the latter, please create a new one

[hannahmadison]

@Sameerlite Will do! Thanks for the quick response!