Bug: readFile fails to interpret PDF content with Google models

[zwpaper]

Bug Report

Describe the bug

The readFile tool is expected to read and interpret the content of PDF files when using Google models. However, when attempting to read test.pdf, the tool returned raw binary data instead of the extracted text content, even when using google/gemini-2.5-pro.

To Reproduce

1. Have a PDF file in the working directory.
2. Use the readFile tool to read the PDF file.
3. Observe the output, which is binary/uninterpreted data.

Expected behavior

The tool should return the human-readable text content extracted from the PDF file.

Actual behavior

The tool returns a binary string representation of the PDF's contents.

Additional context

• This issue was encountered when trying to read the content of test.pdf.
• The model in use was google/gemini-2.5-pro.

🤖 Generated with Pochi

[zhanba]

seems like the gemini model problem, work well with claude model

• gemini

• claude