NEW DeepSeek V4 Pro: Testing Reveals Critical Flaws

Some new AI models are excellent, some have problems. I perform a causal reasoning test to test for my scientific application, with an advanced reasoning complexity. This is not a standard test from the internet, which is known for years. I perform on my own dedicated performance test - for my domain knowledge (science). My Youtube Playlist with all AI models tested (identical test) is available here https://www.youtube.com/playlist?list=PLgy71-0-2-F0Rla8lu5ZldpYQUfXM_5bT Compare with the latest Qwen3.6 A3B performance here https://www.youtube.com/watch?v=Gnk-me1UqfE 00:00 DeepSeek V4 Pro & FLASH 00:23 Live test 02:27 First results 10:25 V4 PRO crash 15:34 V4 PRO 2nd crash 15:41 V4 Flash validation run 17:02 V4 Flash Success Validation #deepseek #deepseekai #deepseek_v4 #aitesting