GPT-5.4 in ChatGPT improves reasoning, coding, tool use, long-context work, and agent-style tasks. Here is what changed and why it matters.
user@cgh.mx:~$ ls -la /content/posts/
Anthropic says Project Glasswing will give critical infrastructure defenders early access to Claude Mythos Preview to find and fix serious vulnerabilities before attackers do.
Claude Code’s auto mode reflects a broader move toward coding agents that can keep working with less step-by-step supervision, but the real question is where autonomy still needs boundaries.
Claude Opus 4.7 is Anthropic’s latest flagship coding model, aimed at harder multi-step engineering work, better visual understanding, and more reliable long-running agent behavior.
Linux 7.0 brings official Rust support, stronger security direction, new filesystem work, and performance improvements that matter beyond the version number.
Gemini CLI now supports subagents, a practical step toward cleaner AI workflows with specialized helpers and less context overload.
GPT-5.4-Cyber is not just another model launch. It shows how OpenAI is creating more permissive, security-focused AI access for verified defenders while keeping tighter controls around higher-risk capabilities.
Google’s upgraded desktop app for Windows is more than a simple launcher. It shows how search is moving out of the browser and toward an always-near desktop layer with AI answers, screen awareness, and visual lookup.
Germany’s StEnSea project is heading toward a California test with a 9-meter concrete sphere, but it is not really a battery in the usual sense. It is underwater energy storage based on pressure and pumped-hydro logic.
Hermes, OpenClaw, and Claude Code overlap around AI agents, but they are built for different jobs. The most useful comparison is not who wins, but what each tool is actually trying to be.