China\'s AI company MiniMax has launched M3.1 Flash, a large language model designed to boost inference speed and sharply cut usage costs. The company said the model reduces AI agent task costs by 70 percent versus its predecessor and increases inference speed about threefold. Built on a mixture-of-experts architecture, it supports text, image and video inputs and up to 1 million tokens of context. MiniMax also disclosed pricing and availability across its API, subscription service and coding app.

U.S. big tech companies are moving into open-weight AI models long led by Chinese firms, putting the competitive landscape in focus. Meta unveiled its 816 billion-parameter Muse Spark Open model, describing it as the best U.S. open-weight model to date. The model supports up to 1 million tokens of context and takes text, images, video and audio as input. Meta outlined local deployment using two Mac Studio systems. Nvidia is also weighing further investment in or an acquisition of ReflectionAI, the Financial Times reported.