Cosmic Rundown: Train Sim Solo Dev, Brain-Targeted AI Video, and Invisible Tools

GPT-5.6 landed on the front page of Hacker News today with 1,448 points and over 1,000 comments. That's the seventh major model release in under two weeks (Sonnet 5, Kimi K2.7, GLM 5.2, Leanstral, Robostral, Grok 4.5, now GPT-5.6). The release cadence is not slowing down.

Here are today's top stories.

GPT-5.6 Is Here

OpenAI's latest drops with strong benchmark numbers and the usual wave of developer experimentation. Notably, the model-churn argument gets stronger every week: if your stack is hard-wired to one model, you are already re-evaluating.

GLM 5.2 Running on a Slow Computer

877 points and climbing. Developers are getting capable open models running on commodity hardware. The local-model category is real and growing fast.

Bun Goes Rust

Bun is rewriting performance-critical internals in Rust. The JS runtime wars are heating up and developers are paying close attention to toolchain foundations.

Write Code Like a Human Will Maintain It

153 points on a piece arguing that AI-generated code needs to be written for the human who maintains it, not just for the model that generates it. The maintainability theme keeps recurring: Godot's AI-code ban, code review philosophy, and now this. A governed, reviewable content and code layer keeps becoming more obviously necessary.

Chat Control Passes in the EU

The EU's Chat Control legislation advanced. Mass scanning of private communications. Significant for anyone building communication tools in Europe.