Just repeated this test.
- Cursor with GPT 5 and Claude Sonnet 4
- Opencode with GPT 5 and Claude Sonnet 4
- Gemini CLI with Gemini Pro 2.5
- Claude Code with Claude Sonnet 4
You probably already know, but Claude Code is the real deal. Don't waste your time with anything else.
Just took another serious attempt on LLMs for solving a non-trivial / not-part-of-the-training-data software engineering task. Tried Copilot Workspaces, GPT-4o with Plandex-ai and Claude Sonnet 3.5 using the web interface and for all three of them the results were pure garbage.
