[Feat] Add support for 0 cost models

[Sameerlite]

Relevant issues

Adding back reverted #19027

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• 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

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 2, 2026 8:01am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR adds support for zero-cost models that bypass budget checks when both input_cost_per_token and output_cost_per_token are explicitly set to 0. This allows on-premises or free models to remain accessible even when users exceed their budget limits.

Key changes:

• Added _is_model_cost_zero() helper function in auth_checks.py that checks if a model has zero cost by querying the router's model group info
• Modified common_checks() to accept a skip_budget_checks parameter that conditionally wraps all budget validation logic (team, organization, user, team member, end-user, and global proxy budgets)
• Integrated zero-cost check in two places in user_api_key_auth.py (lines 608-620 and 1027-1039) before calling common_checks()
• Added comprehensive test suite with 590 lines covering all budget bypass scenarios
• Updated documentation explaining the feature, configuration, and behavior

Implementation approach:
The implementation follows a conservative approach: only models with both costs explicitly set to 0 (not None) will bypass budget checks. The llm_router.get_model_group_info() method is called in the request path to determine model cost status.

Issues found:

• Code duplication: The zero-cost check logic appears twice in user_api_key_auth.py (could be extracted to a helper)
• The comparison input_cost > 0 or output_cost > 0 doesn't handle negative costs (though unlikely in practice)

Confidence Score: 4/5

• This PR is generally safe to merge with minor considerations around performance and code quality
• Score reflects solid implementation with comprehensive tests and documentation, but concerns about code duplication and potential performance impact from calling llm_router.get_model_group_info() in the critical request path. The logic is sound and conservative (requiring explicit zero costs), but the duplication should be addressed and performance should be monitored.
• Pay attention to litellm/proxy/auth/user_api_key_auth.py for code duplication and performance monitoring of the router call in the request path

Important Files Changed

Filename	Overview
docs/my-website/docs/proxy/custom_pricing.md	Added comprehensive documentation for zero-cost models feature, explaining configuration and behavior
litellm/proxy/auth/auth_checks.py	Adds _is_model_cost_zero helper and wraps budget checks in conditional block based on skip_budget_checks flag
litellm/proxy/auth/user_api_key_auth.py	Integrates zero-cost check in two authentication paths with duplicated code blocks; wraps budget checks conditionally
tests/proxy_unit_tests/test_zero_cost_model_budget_bypass.py	Comprehensive test suite covering zero-cost model scenarios across user, team, end-user budgets, and edge cases

Sequence Diagram

sequenceDiagram
    participant C as Client
    participant U as user_api_key_auth
    participant A as auth_checks
    participant R as Router
    participant CC as common_checks

    C->>U: Request
    U->>U: Extract model
    U->>A: Check cost
    A->>R: Get pricing
    R-->>A: Pricing data
    
    alt Zero cost
        A-->>U: Bypass enabled
    else Has cost
        A-->>U: Normal flow
    end
    
    U->>CC: Common checks
    
    alt Normal flow
        CC->>CC: All budget checks
    else Bypass
        CC->>CC: Skip budget checks
    end
    
    CC-->>U: Result
    U-->>C: Response

Loading

[greptile-apps]

_{4 files reviewed, 3 comments}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

This code block for checking zero-cost models is duplicated at lines 1027-1039. Consider extracting this logic into a helper function to reduce duplication and maintain consistency.

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}

Prompt To Fix With AI

This is a comment left during a code review.
Path: litellm/proxy/auth/user_api_key_auth.py
Line: 608:620

Comment:
This code block for checking zero-cost models is duplicated at lines 1027-1039. Consider extracting this logic into a helper function to reduce duplication and maintain consistency.

<sub>Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!</sub>

How can I resolve this? If you propose a fix, please make it concise.

[greptile-apps]

Verify that llm_router.get_model_group_info() uses caching and doesn't cause performance degradation in the request path, as per the custom rule about avoiding expensive operations.

Context Used: Rule from dashboard - What: Avoid creating new database requests or Router objects in the critical request path.

Why: Cre... (source)

Prompt To Fix With AI

This is a comment left during a code review.
Path: litellm/proxy/auth/auth_checks.py
Line: 103:103

Comment:
Verify that `llm_router.get_model_group_info()` uses caching and doesn't cause performance degradation in the request path, as per the custom rule about avoiding expensive operations.

**Context Used:** Rule from `dashboard` - What: Avoid creating new database requests or Router objects in the critical request path.

Why: Cre... ([source](https://app.greptile.com/review/custom-context?memory=0c2a17ad-5f29-423f-a48b-371852ac4169))

How can I resolve this? If you propose a fix, please make it concise.

[greptile-apps]

The condition input_cost > 0 or output_cost > 0 doesn't handle negative costs (though unlikely in practice). Consider using input_cost != 0 or output_cost != 0 for completeness, or add validation that costs cannot be negative.

_{Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!}

Prompt To Fix With AI

This is a comment left during a code review.
Path: litellm/proxy/auth/auth_checks.py
Line: 126:126

Comment:
The condition `input_cost > 0 or output_cost > 0` doesn't handle negative costs (though unlikely in practice). Consider using `input_cost != 0 or output_cost != 0` for completeness, or add validation that costs cannot be negative.

<sub>Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!</sub>

How can I resolve this? If you propose a fix, please make it concise.