Positive feelings about AI outweigh negative ones in 34 of 37 countries surveyed with Microsoft, but only 36% of Americans expect it to help their country and 74% say it worries them.
The first Claude since the “pace the frontier” essay is cheaper and faster, and by Anthropic’s own account it still tried to slip its sandbox in 1.5% of test runs.
Sol lists at $2 per million input tokens and Luna at 10 cents. The release came 90 minutes after Anthropic’s Opus 5.5, and the price war has moved from the frontier to the models most developers use.
Announced on the sidelines of the UN General Assembly, the deal puts OpenAI’s vulnerability-hunting models in the hands of a country whose national CERT handled nearly 6,000 incidents in 2025.
A six-author paper shows the Mathieu group M23 is a Galois group over the rationals, closing a list open since the 1980s. The acknowledgments credit Claude and ChatGPT; the text, the authors say, is all human.
SpaceXAI's new model, released on Monday, beats its predecessor on every coding benchmark the company lists and keeps the $2 and $6 per-million pricing. Independent testing finds it needs about 81,000 output tokens per task, more than double Grok 4.6.
The province and the local school district filed in federal court in San Francisco on Monday, alleging OpenAI's own reviewers flagged the shooter's account in June 2025 and the company never told police.
A Monday post from OpenAI says fully autonomous recursive self-improvement "is not happening today" and should not be pursued until it can be done safely. It wants US-led technical standards, but not licences.
At its Apsara Conference in Hangzhou on Tuesday, Alibaba showed the Zhenwu V900 accelerator, set a 20-gigawatt capacity target for 2032 and said its next Qwen generations will be two to four times the size of the current flagship.
The bills signed on Monday create separate electricity tariffs for loads of 75 megawatts and more, require water-use disclosure before local approval, and pull data centres out of blanket environmental-review exemptions.
The 1.02-trillion-parameter Pro and 309-billion-parameter Flash models arrived on Monday with their reinforcement-learning code and thousands of training environments, and score level with Grok 4.7 on one independent index.