Deepseek V4 pro vs Minimax M3. Judge is Opus 4.8. Results are disappointing

built an app. Used GLM 5.2 to make a Build Plan. Asked deepseek v4 pro and minimax m3 to implement the build plan faithfully (two separate projects, used Opencode desktop). Then asked claude code (Opud 4.8 high) to audit both implementations. Results are in the image. Very disappointed with deepseek, lot of issues. Totally surprised my M3 which cost me the same miniscule amount as deepseek (around 35 to 40 cents), but with mostly solid results. Surprised bcos last time when i used M2.7 for coding, it was th