Anthropic accuses Alibaba of copying Claude; 'tokenmaxxing' fades

Anthropic told the US Senate Banking Committee, in a June 10 letter, that operators tied to Alibaba and its Qwen AI lab ran roughly 25,000 fraudulent accounts to query Claude more than 28.8 million times between April 22 and June 5 — what Anthropic calls the largest known "distillation" attack against it, a technique that trains a cheaper rival model on a stronger one's outputs. It echoes earlier campaigns Anthropic attributed to DeepSeek, Moonshot, and MiniMax, and sharpens the US-China fight over who gets to copy frontier AI.

For two years, companies pushed staff to burn as many AI tokens as possible, treating heavy usage as a proxy for innovation. That is reversing: enterprises now want clear return on investment and lower-cost models, after cases like Uber blowing its annual AI budget in four months and startup Lindy moving all of its traffic from Claude to China's far cheaper DeepSeek. The average cost per million tokens fell from about ten dollars to two dollars and fifty cents in a year, pressuring OpenAI and Anthropic to defend premium pricing as cheaper models close the quality gap.

Baseten, which runs other companies' AI models in production, raised one and a half billion dollars at a valuation of up to thirteen billion dollars — a 160% jump in five months and one of the largest AI-infrastructure rounds on record. The raise signals that "inference," the cost of actually running a model to answer each query, has become a contested category of its own as open and closed models converge in capability. Baseten says it now handles more than one billion inference calls a day, with revenue up roughly twenty times year over year.

General Intuition, a spinout of gameplay-clip site Medal, raised three hundred twenty million dollars, led by Khosla Ventures, at a valuation of 2.3 billion dollars. The startup turns hundreds of millions of hours of recorded gameplay — complete with the exact buttons players pressed and when — into training data for "world models" that learn physics and cause-and-effect. The goal is AI agents that can act in the real world, from robots to assistants, by first mastering the simulated one.

Fintech Airwallex raised three hundred twenty million dollars in Series H funding at a valuation of eleven billion dollars, pitching itself as the financial infrastructure for a coming wave of AI agents that handle business payments, expenses, and treasury on their own. The bet: as companies hand routine money tasks to software agents, those agents still need bank accounts, cards, and cross-border rails — and Airwallex wants to be the layer they run on. It is one of several June raises wagering that agents will soon move real money.