fix(moonshot): preserve image_url blocks in multimodal messages

[Chesars]

Relevant issues

Fixes #20862

Pre-Submission checklist

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

Changes

Moonshot's _transform_messages unconditionally flattened content arrays to plain strings via handle_messages_with_content_list_to_str_conversion(), dropping all non-text content blocks (image_url, input_audio, video_url, file, etc.). Multimodal models like kimi-k2.5 accept the standard OpenAI content array format.

Fix: Before flattening, check if any message contains a non-text content part (type != "text"). If so, skip flattening and preserve the content array. If all parts are text-only, flatten as before (required by Moonshot text models). This covers the full OpenAI content spec, not just images.

Example request

import litellm

response = litellm.completion(
    model="moonshot/kimi-k2.5",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "What is in this image?"},
                {
                    "type": "image_url",
                    "image_url": {"url": "https://example.com/image.png"},
                },
            ],
        }
    ],
)

Before fix: the content array was flattened to the string "What is in this image?" — the image_url block was dropped.

After fix: the content array is preserved as-is, and the multimodal payload reaches the Moonshot API.

Files changed

• litellm/llms/moonshot/chat/transformation.py — Skip flattening when any content part has type != "text"
• tests/test_litellm/llms/moonshot/test_moonshot_chat_transformation.py — 3 new tests: image_url preserved, input_audio preserved, text-only flattened
• docs/my-website/docs/providers/moonshot.md — Add "Image / Vision Support" section with usage example

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 19, 2026 7:46pm

[greptile-apps]

Greptile Summary

This PR fixes a bug where Moonshot's _transform_messages was unconditionally flattening content arrays to strings, causing image_url blocks to be silently dropped for vision models like kimi-k2.5.

Key changes:

• Added detection logic to check if any message contains an image_url block before flattening
• If images are present, content arrays are preserved; otherwise, they're flattened to strings as required by text-only models
• Pattern mirrors the Mistral provider implementation (lines 250-259 in litellm/llms/mistral/chat/transformation.py)
• Two new tests validate both the image preservation and text flattening behavior
• Documentation updated with vision support section and usage example

The fix is minimal, focused, and maintains backward compatibility for text-only requests while enabling vision capabilities.

Confidence Score: 5/5

• This PR is safe to merge with no issues found
• The fix is minimal, well-tested, follows the exact pattern used by Mistral provider, and addresses a specific bug where image_url blocks were being dropped. Tests cover both the positive case (image preservation) and negative case (text flattening).
• No files require special attention

Important Files Changed

Filename	Overview
litellm/llms/moonshot/chat/transformation.py	Adds image_url detection logic to preserve content arrays for vision models while maintaining text-only flattening for other models
tests/test_litellm/llms/moonshot/test_moonshot_chat_transformation.py	Adds two comprehensive tests for image_url preservation and text-only flattening behavior
docs/my-website/docs/providers/moonshot.md	Adds clear documentation section for vision support with usage example

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User calls litellm.completion] --> B[MoonshotChatConfig._transform_messages]
    B --> C{Check all messages for image_url}
    C -->|Has image_url| D[Skip flattening]
    C -->|No image_url| E[Call handle_messages_with_content_list_to_str_conversion]
    D --> F[Content array preserved]
    E --> G[Content flattened to string]
    F --> H[Send to Moonshot Vision API]
    G --> I[Send to Moonshot Text API]
    H --> J[Image reaches API successfully]
    I --> K[Text-only request processed]

Loading

_{Last reviewed commit: c5cec60}

[greptile-apps]

_{3 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}