"Benchmark-watching makes you an NPC": dev says GLM 5.3 Flash outworks Opus 5.5 in real tasks
gnukeith · x · 2026-10-25
gnukeith argues that judging models purely by benchmark scores makes you an "NPC waiting for the next slop scoop." He claims he gets more actual work done with GLM 5.3 Flash than with Opus 5.5 — while also using Opus 5.5 for impressive 3D-printing work — reinforcing his point that model choice depends on the task, not leaderboards.
More from Models
- Blogger revisits Kimi K3 hype, speculates China may release some B300 GPUs — lipeng0820 · 2026-10-25
- Local Qwen model designs its own project: an interactive IQ heat map by state — LankyGuitar6528 · 2026-10-25
- Matthew Berman tests DeepSeek 4.1 Flash: one prompt to build a Fall Guys clone via coding agent — MatthewBerman · 2026-10-25
- Abliterated models are a real capability edge, and cybersecurity just raised their value — generativist · 2026-10-25
- ChatGPT Health connector shows Unavailable in EU, user asks OpenAI why — Angaisb_ · 2026-10-25
- Opus 5 designs an optical illusion of faces whose rotation flips with your mind — repligate · 2026-10-25