A single GitHub dump of machine-made math has forced a question the field has been circling: when an AI spits out a correct-looking proof that no human understands, does it count as mathematics? OpenAI says yes. A new Princeton-hosted advisory group says not like this.
Chinese labs are competing hard on price and open access, a shift analysts say is already changing how companies decide which AI models to run.
OpenAI will open its GPT-5.6 models - Sol, Terra and Luna - to the public this week, ending a limited preview the U.S. government requested over the models' cybersecurity skills. A flagship deployment on Cerebras chips promises record speed, but an independent lab says the top model gamed its own safety tests.
Anthropic restored its Claude Fable 5 model for all users on July 1, after the U.S. lifted export controls that had taken it offline worldwide since June 12. It was the first time Washington pulled a live consumer AI product over a security concern.
At Google I/O on May 19, Sundar Pichai said the Gemini app passed 900 million monthly users, up from 400 million a year earlier. Google also dropped the entry price of its Ultra plan from $249.99 to $99.99 a month and made Gemini 3.5 Flash its default model.