fix(utils): normalize camelCase thinking param keys to snake_case

[Chesars]

Relevant issues

Fixes clients (e.g. OpenCode's @ai-sdk/openai-compatible) sending {"thinking": {"type": "enabled", "budgetTokens": 32000}} which causes thinking.enabled.budget_tokens: Field required validation errors.
Reference: anomalyco/opencode#10109

Pre-Submission checklist

• 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
• 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

Clients like OpenCode's @ai-sdk/openai-compatible send budgetTokens (camelCase) instead of budget_tokens (snake_case) in the thinking parameter. LiteLLM's Anthropic provider validation rejects this with a Field required error.

This PR adds early normalization of camelCase keys in the thinking param to snake_case before reaching the Anthropic API, following the same pattern as validate_and_fix_openai_tools().

import litellm

response = litellm.completion(
    model="anthropic/claude-sonnet-4-20250514",
    messages=[{"role": "user", "content": "What is 2+2? Reply in one word."}],
    thinking={"type": "enabled", "budgetTokens": 5000},  # camelCase from client
)

Without the fix: thinking.enabled.budget_tokens: Field required
With the fix: works correctly, budgetTokens is normalized to budget_tokens.

Files changed

• litellm/utils.py — Added validate_and_fix_thinking_param() that converts budgetTokens → budget_tokens
• litellm/main.py — Imported and called the new function in the validation section of completion()
• tests/test_litellm/test_utils.py — 5 unit tests: None input, already-snake_case, camelCase normalization, both-keys-present, no-mutation of original dict

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 21, 2026 0:37am

[greptile-apps]

Greptile Summary

This PR fixes a validation error (thinking.enabled.budget_tokens: Field required) caused by clients (e.g., OpenCode's @ai-sdk/openai-compatible) sending budgetTokens (camelCase) instead of budget_tokens (snake_case) in the thinking parameter.

• Adds validate_and_fix_thinking_param() in litellm/utils.py that normalizes camelCase keys to snake_case before the param reaches provider-specific validation
• Calls the new function in the validation section of completion() in litellm/main.py, following the existing pattern of validate_and_fix_openai_tools() and validate_and_fix_openai_messages()
• Includes 5 unit tests covering None input, already-correct input, camelCase normalization, both-keys-present precedence, and non-mutation of original dict
• The acompletion() async path is also covered since it delegates to completion() internally

Confidence Score: 5/5

• This PR is safe to merge — it adds a small, well-tested normalization function that handles a clear client compatibility issue.
• The change is minimal and focused: a single normalization function that converts one camelCase key to snake_case. It follows existing validation patterns in the codebase, makes a defensive copy to avoid mutation, handles all edge cases correctly, and includes 5 unit tests with no network calls. The function is placed early in the validation pipeline and covers both sync and async code paths.
• No files require special attention.

Important Files Changed

Filename	Overview
litellm/utils.py	Adds validate_and_fix_thinking_param() that normalizes camelCase budgetTokens to snake_case budget_tokens. Clean implementation with proper shallow copy and correct handling of edge cases.
litellm/main.py	Imports and calls the new validation function in the correct validation section of completion(), following existing patterns.
tests/test_litellm/test_utils.py	Adds 5 pure unit tests with good coverage of edge cases. No network calls — all tests are mock-free and test the function directly.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["Client sends thinking param\n(camelCase or snake_case)"] --> B["completion() in main.py"]
    B --> C["validate_and_fix_thinking_param()"]
    C --> D{thinking is None\nor not dict?}
    D -- Yes --> E["Return as-is"]
    D -- No --> F{Has budgetTokens?}
    F -- No --> G["Return copy unchanged"]
    F -- Yes --> H{Also has budget_tokens?}
    H -- No --> I["Rename budgetTokens → budget_tokens"]
    H -- Yes --> J["Drop budgetTokens, keep budget_tokens"]
    I --> K["Return normalized dict"]
    J --> K
    G --> K
    K --> L["Continue to provider-specific\ntransformation in llms/"]

Loading

_{Last reviewed commit: 599d52b}

[greptile-apps]

_{3 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}