Hacker Newsroom for 25 July recaps major Hacker News stories, moving through claude opus 5, losing focus, software gets worse, open weight rules. 1. Claude Opus 5 The next story is Anthropic’s launch of Claude Opus 5, a new model the article says gets close to Claude Fable 5’s frontier-level intelligence at about half the cost, with strong results on coding, automation, science, and visual tasks. The post frames Opus 5 as a practical everyday flagship, pointing to benchmark wins over Opus 4. Story link Hacker News discussion 2. Losing Focus The next story is It's getting harder to focus every day, a post about how one developer feels his ability to do deep work slipping even when he wants to code, study, or write. In the post, he traces that slide from intentional computing toward constant escape hatches like chat, feeds, and especially LLM side quests, where outsourcing a task still leaves his attention split across multiple half-finished threads. Story link Hacker News discussion 3. Software Gets Worse The next story asks why software keeps getting worse even as AI tools promise to make coding easier. In this post, Piotr Chmielowski argues that the real problem is not AI itself but growing software complexity, feature pressure, and companies that reward visible output over boring stability work. Story link Hacker News discussion 4. Open Weight Rules The next story is about Nvidia, Microsoft, Meta, and more than 20 other companies urging policymakers not to impose what they call premature restrictions on open-weight AI models, arguing in this CNBC article that downloadable models keep competition alive and spread AI capability beyond a handful of closed labs. The news story ties that push to the rise of Chinese models like Kimi K3, the debate over sanctions and distillation, and the notable absence of OpenAI and Anthropic from the letter even though Sam Altman later said the U. Story link Hacker News discussion 5. Flux 3 The next story is Flux 3, a Black Forest Labs post introducing a new multimodal model that learns from images, video, and audio together, with the company pitching it as a step toward systems that can generate media, predict actions, and eventually support physical AI. The post claims strong early results in text-to-video, image generation, multilingual audio, and robotics-related action prediction, plus an eventual open-weight backbone, which matters because it frames the next AI race as unified world models rather than separate tools for text, images, and video. Story link Hacker News discussion 6. Camera Token Leak The next story is a post about a Hanwha security camera firmware image that exposed a GitHub admin token inside the files behind the device’s login page. By unpacking the firmware, the post shows how a frontend build appears to have dumped an entire CI environment into shipped assets, leaking a token with admin access to hundreds of repositories and surfacing some strange internal addresses tied to Department of Defense IP space. Story link Hacker News discussion That's it for today, I hope this is going to help you build some cool things.