Perplexity details its two-stage training process for web search agents, showing its open-source Qwen-based model outperforms GPT-5.4 and Claude Sonnet on accuracy while costing a fraction of the price.
Vercel's investigation into a recent security breach found the attack was more widespread than initially reported, involving malware distribution and uncovering older, unrelated account compromises.
Ideogram's new Custom Branded Models feature lets companies upload 15-100 images to train a dedicated AI model that generates consistent, on-brand visuals.
Xiaomi's new MiMo V2.5 AI models are now in public testing, with the top model targeting Claude Opus and GPT-5.4 performance. It aced complex coding tasks and will soon be open-sourced globally.
Billionaire Tianqiao Chen argues that building a global AI company requires more than just moving headquarters—it needs compliance designed in from the start.
Japanet Holdings, a Japanese TV shopping company, quadrupled its venture fund to $200 million after early bets on Anthropic and xAI turned into paper fortunes.
NeoCognition emerges from stealth with $40M to build AI agents that learn on the job, targeting the reliability problem that plagues today's AI assistants.
On a podcast, OpenAI's Greg Brockman revealed that GPT-5.4 Pro has solved a new Erdős problem, and teased an upcoming model to fix AI's writing shortcomings.
Sam Altman reveals the moment Elon Musk demanded 'absolute control' of OpenAI, breaking their partnership. The trial is set for April 27th.
OpenAI president Greg Brockman has taken over product integration, narrowing the company's focus to three lines — and Sora is out.
