DeepSeek-V3/R1: MLA 93.3% compression breaks the long-context VRAM wall 99.8k Linux 6.12 io_uring: SendZc zero-copy pushes 4.5M QPS on one core 88.4k React 19 & Next 15: React Compiler auto-memoization and streaming PPR 92.1k PostgreSQL 17 pgvector: AVX-512 vector search 400% faster under load 81.5k Cilium & eBPF: Socket-level bypass redirect skips the L4 stack 76.9k ML-KEM (Kyber-768): NIST post-quantum crypto lands in hybrid TLS 1.3 65.2k Python 3.13 No-GIL: Free-threading brings true multi-core parallelism 94.3k WebGPU WGSL Compute: Ten million SPH particles at 60 FPS in the browser 79.6k