Repository navigation
OpenCode doesn't perform as well as Claude Code when processing large files #2003
Description
First, I used the same model, qwen3-coder.
I used the same prompts and ran the code multiple times to minimize randomness.
I have a C++ file with nearly 20,000 lines. Let's not discuss whether this is a good design, but it exists.
- Claude Code tends to use the Search command to locate a precise location before partially reading the code. While OpenCode also partially reads the code, it doesn't actively use Search (strangely, given that it uses the same model), instead reading the file incrementally from the beginning.
- When Claude Code reads a file large enough to break the context window (large enough to break the context window), it directly reports an error instead of sending it to the large model. OpenCode also encountered a situation where the context window was exceeded during execution.
I'm a fan of both OpenCode and @thdxr , and I hope they continue to iterate towards perfection.
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Aug 16, 2025
This issue might be a duplicate of existing issues. Please check:
Feel free to ignore if none of these address your specific case.
rekram1-node commentedon Aug 16, 2025
yeah large file support needs some work
hujianxin commentedon Aug 17, 2025
@rekram1-node Thanks. Let me add to my description. When Claude Code tries to read a large file, it finds that the file is too big. Then, when you send this to the LLM, it will inform the large model of the fact that the file is too large. After that, it keeps the conversation going through precise searching. This seems very smooth and AGI-like.
rekram1-node commentedon Aug 17, 2025
thank you @hujianxin