Just say no to OpenAI and Anthropic

Blog post by Micah Lee, who describes himself in it as “a software and security engineer”, page date 2026-10-11. He argues against relying on OpenAI and Anthropic on political, privacy and access grounds. He objects that both labs decide who gets less restricted cyber models (OpenAI’s Trusted Access for Cyber, Anthropic’s Cyber Verification Program) and says of open-weight models: “So far, open weight models seem to be about four months behind the fancy proprietary models from OpenAI and Anthropic.” His “vibes-based opinion” is that GLM-5.3 is “basically on par with Claude Opus 5.5 for software engineering”; this is his impression, not a measurement. He recommends not sending data to Z.ai itself (he cites the ZCode source-upload incident), and using secure-enclave inference providers (Tinfoil, Maple AI) or routers such as OpenCode Go; he quotes Tinfoil’s GLM-5.3 price as $1.80 input, $0.80 cached and $5.75 output per 1M tokens (not checked here). He also mentions Mistral Large 4 as the one non-US, non-Chinese option.

Hacker News: 157 points and about 149 comments on 2026-10-13 (https://news.ycombinator.com/item?id=50060500). Read on 2026-10-13.