What's trending in AI right now, from the app charts to the community feeds. Real links, our take.
Anthropic released Claude Opus 5
The new Claude Opus 5 landed on July 24 and immediately dominated AI video channels, pulling several thousand views per hour in the first day. It is Anthropic's new flagship model, priced the same as its predecessor but scoring significantly higher on coding and reasoning benchmarks. Claude Max and Claude Pro subscribers are already on it by default. The broader reaction in developer communities has been that it represents a genuine step up, not just a version bump.
Google's Gemini Spark is now rolling out
Google opened Gemini Spark to US Google AI Pro subscribers this week, included in the existing subscription at no extra charge. It sits between the free Gemini tier and the premium AI Ultra plan, designed for the everyday stuff: drafts, research, coding help. If you are already paying for Google AI Pro, you can select it from the model menu today. The rollout pushed "gemini" into the US search trends alongside the Claude Opus 5 launch, so both companies are in the news for the same reason right now.
ChatGPT had a difficult 24 hours
Early this morning, ChatGPT went down worldwide for about 50 minutes, taking Codex and 12 API endpoints with it before service recovered around 6 AM ET. It was brief, but it sent "ChatGPT" to the top of search trends in both the US and UK simultaneously, which tells you something about how central it has become to people's daily routines.
Separately, a Florida man sued OpenAI this week alleging that ChatGPT-4o dismissed warning signs that preceded a near-fatal pulmonary embolism in 2025, discouraging him from seeking medical care. Legal observers are calling it the first case arguing that a general-purpose chatbot can be held liable for dangerous medical advice. OpenAI says the service is not designed to replace a doctor.
Cursor rebuilt SQLite from its own manual
Cursor published results this week showing that a swarm of AI agents rebuilt SQLite in Rust using only the 835-page SQLite documentation, with no access to the original source code. The finished implementation passed 100% of the standard SQL test suite and cost about $1,300 using a planner-plus-cheaper-worker setup. The experiment circulated widely in developer feeds because it shows multi-agent systems handling the kind of long-horizon, spec-driven engineering work that has until recently required months of human effort.
An AI found Redis zero-days in 27 minutes
Researchers using Moonshot AI's Kimi K3 model discovered 19 zero-day vulnerabilities in Redis in about 90 minutes, with one agent producing a working remote code execution exploit in just 27 minutes. Redis shipped seven security patches on July 23 in response. Security researchers are watching this closely because it is one of the clearest examples yet of an AI system doing autonomous vulnerability research from start to finish, including building a working exploit, rather than assisting a human analyst who is already leading the work.