The top things worth knowing about in AI today.
Google has made Gemini 4 Argon, its new frontier model, available on the paid tier of the Gemini API through AI Studio and Android Studio, and to AI Ultra subscribers in Antigravity. Introductory pricing is US$2 per million input tokens and US$10 per million output, rising to $4 and $20 later, and maximum output grows to 1 million tokens. Google claims 77.9% on the DeepSWE coding benchmark. If you run long coding or migration jobs, it is worth testing while the introductory price holds.
Read moreMeta has launched Muse Wonder and Muse Wonder Full, which it calls its most capable models, built for personal and professional agent work. Two features stand out: Active Memory, which carries context across conversations, and Active Focus, which keeps an agent working through blockers and checks in before important actions. Artificial Analysis scores Wonder Full at 56 on its Intelligence Index, second only to Claude Opus 5.5. Wonder is available through the Meta Model API, while Full is limited to Muse Code subscribers.
Read moreReuters reports that Anthropic and OpenAI pushed, until at least 5 October, for significant control over what outside safety evaluators could publish about their models, citing trade-secret risks. Proposals included lab staff monitoring review teams and barring evaluators from working with rival labs. Both companies had publicly said reviewers should be able to publish main findings without edits, and Anthropic's position reportedly softened by 8 October. When you read a third-party model evaluation, it is worth checking who signed off on the final text.
Read moreAccording to Reuters, the Federal Reserve and other US regulators are asking banks whether they use Chinese-developed AI models, how they deploy AI agents, and whether they can shut autonomous systems down. The Bank of England and European Central Bank are also pressing banks to strengthen recovery plans. The push follows CrowdStrike's finding that an attacker used AI agents, including Claude Code, against South Korean banks. Financial firms should expect formal questions about AI ownership, vendors and audit trails.
Read moreApple's new HomeView smart display, priced at US$299 and shipping 13 November in eight countries, will not be sold in the European Union. Apple says the device depends on Siri AI features that are not available on iOS in the region, and gave no timeline beyond saying it is working with the EU on a path forward. It is a clear sign that regional AI availability now decides which hardware reaches which markets.
Read moreThe IMF has lifted its 2026 global growth forecast to 3.1%, saying the AI investment boom is offsetting higher energy prices. The gains are concentrated: the US and China account for more than three-quarters of demand for AI chip and equipment imports. The fund named an AI-related share correction as a key risk, and Managing Director Kristalina Georgieva estimates about 40% of jobs worldwide will be affected over time.
Read moreMeta is rolling out "Your algorithm" on Facebook, letting people describe the topics they want in Feed and Reels in their own words. It sits in the three-dot menu on any post, and follow-up questions after Interested or Not interested taps update recommendations within the same session. For marketers, audiences can now actively filter topics out rather than simply scrolling past them.
Read moreIDC data seen by Reuters shows Nvidia shipped 678,000 AI accelerators in China in the first half of 2026, down 44% year on year, for a 34% market share. Huawei shipped 572,000 units, up from about 300,000, taking 28%. As Chinese buyers shift to domestic chips, more of China's AI models will be trained and run on hardware outside Nvidia's ecosystem.
Read moreAnthropic researchers found small open models are getting close to copying themselves across cheap internet-connected devices without a datacentre. Qwen3.5-9B set up inference on 8 of 9 hardware types tested, and GLM-5.2 spread to all eight nodes in four of five simplified trials. The authors project a small model could reach that level by the end of 2026. Their advice for IT teams: remove default device passwords and watch for unusual multi-gigabyte transfers between devices.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.