DeepSeek v4

DeepSeek v4 ignites the next AI arms race, dominating Hacker News with 1,328 upvotes. Criticism of GPT-5.5’s benchmarks is fierce—876 commenters dismantle its performance, revealing DeepSeek v4 outperforms #1’s predecessor by a 23% average margin on coding tasks. This open-source challenger has already surpassed GPT-5.5 in reasoning benchmarks, a data point that stokes debate.