April 2026

Honor’s bright-red Lightning bot stunned crowds at Beijing’s half-marathon on Sunday, crossing the 21km finish line in 50:26 — nearly seven minutes faster than the human world record set by Jacob Kiplimo. Over 100 Chinese-made humanoids raced in a dedicated lane alongside 12,000 human runners, marking just the second year robots joined the event. This […]

The breakthrough behind Canva’s new “Design AI Model” lies in training on the sequence of actions, not just finished designs. Instead of learning only from static images, the model ingests millions of real designs along with every click, layout shift, and style change made by users, capturing how people actually arrive at good work over

Anthropic just launched Claude Opus 4.7, their newest AI brain that's better than rivals like GPT-5.4 and Gemini 3.1 at tough coding tasks. It scores 64.3% on a key test called SWE-bench Pro—up from the old version's 53.4%. But their super-secret model, Mythos Preview, still crushes it at 77.8%. Pricing stays the same, though it

OpenAI has transformed Codex from a mere coding agent into a powerhouse superapp, integrating ChatGPT, Atlas, and advanced tools. The latest update introduces background computer use, allowing Codex to independently operate Mac apps — even those without APIs—while running multiple parallel agents simultaneously. This marks a shift toward seamless, autonomous AI assistance beyond traditional coding.

OpenAI's GPT-5.4-Cyber is a smart AI tool built for cyber defense. It spots hidden flaws in software, like malware or weak spots, by doing tough jobs such as binary reverse engineering. This means it checks compiled code without needing the original plans, helping security experts find dangers fast and fix them before harm happens. Now,

OpenAI's GPT-5.4-Cyber and Anthropic's Mythos mark a pivotal clash in AI-driven cybersecurity, with each fine-tuned for defensive prowess but differing sharply in scope and access. GPT-5.4-Cyber excels in binary reverse engineering, allowing analysts to dissect compiled software for malware, vulnerabilities, and robustness without source code, alongside lowered refusal boundaries for cyber workflows. Mythos, conversely, shines

For the first time since 2012, Google faces losing its crown in global digital advertising. A new forecast reveals Meta's net ad revenue will hit $243.5 billion in 2026, narrowly surpassing Google's $239.5 billion. This marks the end of Google's 14-year dominance, driven by Meta's explosive growth in social platforms like Facebook and Instagram. Meta's

Last week, OpenAI released a 13-page policy paper warning that AI could change society faster than we can handle. It calls for new safety nets, taxes on AI profits, and ways to share wealth as jobs vanish. Days later, a 20-year-old threw a Molotov cocktail at Sam Altman's San Francisco home. The suspect, Daniel Moreno-Gama,

Researchers at Oxford University have built an amazing AI tool that checks heart fat in regular CT scans. It spots tiny changes that doctors can't see with their eyes. This flags people at high risk of heart failure up to five years before it happens. Tested on 72,000 patients, it works 86% of the time.

Perplexity is ditching the Google chase. Its February Computer launch isn't just another search toy—it's a full-blown personal finance powerhouse, plugging straight into your bank accounts via Plaid's 12,000+ network. Pull checking, credit, loans, and brokerage data for read-only access, then watch AI agents spit out budgets, net worth trackers, debt plans, and retirement dashboards

Last summer, Mark Zuckerberg handed Alexandr Wang—fresh from a $14.3B Scale AI acquisition—the reins to Meta Superintelligence Labs, complete with poached talent from top rivals and a blank-check budget. Nine months later, the first fruit: Muse Spark, a proprietary multimodal model handling voice, text, and images with a "contemplating mode" where AI agents debate tough