In a technology reconstruction project, the author conducted in-depth testing of cutting-edge AI models including Deepseek, MiniMax M2.1, and Claude Opus 4.5 to evaluate their performance in real-world development scenarios. The article details the testing environment, including the use of Claude Code v2.0.74 with thinking mode disabled, and model mappings such as the pairing of Opus 4.5 with deepseek-reasoner. The author also shares development experience with multi-channel intelligent scheduling functionality, solving issues with upstream unavailability. By comparing responses from different models, the article reveals the strengths and weaknesses of domestic AI models versus international ones, providing valuable practical case studies for AI developers. Finally, the author mentions publishing three specific responses for community review, demonstrating the spirit of open-source collaboration.
