Moonshot AI has set a target of generating $2 billion in annual revenue by the end of 2026, doubling its reported August run rate. This aggressive financial goal follows the release of its K3 open-weight model earlier this summer, which currently drives approximately 300 billion tokens daily through third-party platforms like OpenRouter. Despite a slight dip in recent usage figures, the projection highlights significant commercial potential for open models, even as Moonshot operates with thinner margins than closed-weight rivals such as OpenAI and Anthropic. The company’s total projected income remains far below those giants, yet the move signals that open weights can still generate substantial returns.
The ambition stands in contrast to ongoing legal disputes regarding Moonshot’s data collection practices. Anthropic alleges the company routed nearly 300,000 requests directly to its Claude Opus model to train Kimi, collecting over 23 million responses in the process. This accusation suggests the lab may have engaged in model distillation that borders on illegality. The situation underscores the tension between rapid commercial scaling and the ethical boundaries of training data acquisition.
- K3 usage on OpenRouter hit 300 billion tokens per day
- Anthropic claims 23 million responses were scraped for training
- Projected revenue would double the August run rate
