Closed-model access risk pushes builders toward open and local

The Samsung TV privacy thread turned into a vendor-risk thread. Commenters said OpenAI reportedly banned the researcher for reverse engineering, and the replies went straight to the stakes: 'What is next? Power and water company suspending your account if your home and car are not streaming enough telemetry.' Others pointed out that off-the-shelf Chinese models like GLM already do plenty of reverse engineering and hacking work. In a separate thread, Anthropic told Claude Code subscribers they can no longer use OpenClaw on their subscription. It offered a one-time credit equal to a month's fee. Users called it a money grab and asked why heavy use of paid token capacity counts as 'outsized strain.'

The local-model thread, 'Has anyone replaced Claude/GPT with a local model for daily coding?', shows where people are heading. One user reports Qwen 3.6 27B dense is roughly equal to Claude Haiku 4.5, maybe Sonnet depending on the task. Others run DeepSeek V4 Flash on 2x RTX Pro 6000 Blackwell at 160 tok/s. The complaints are still real: slower tokens per second on an Apple M4, and a lack of tooling to choose and run models.

The pattern here: terms of service and rate limits are now a product risk, not fine print. Every access cut gives open weights a stronger pitch.