Moonshot AI, one of China’s leading artificial intelligence laboratories, is optimistic about transforming its popular open-weight model into substantial sales growth. According to a report from Bloomberg on Friday, the lab aims to achieve an annual revenue of $2 billion by year-end, which would be double the company’s revenue run rate recorded in August. This ambitious target is a reflection of the impressive success of its K3 model since its launch earlier this summer.
Despite a slight decline in usage statistics for K3 in recent months, current data from OpenRouter indicates that the system’s K3 models are generating up to 300 billion tokens daily.
However, Moonshot’s anticipated revenue still pales in comparison to industry giants like OpenAI and Anthropic, which have been reported at $40 billion and $65 billion in revenue, respectively. As Moonshot’s model weights are publicly accessible, their margins are significantly slimmer than those of closed-weight rivals. Nevertheless, the increasing revenue projections signal that there is still potential for profit from open-weight AI models, even if they don’t match the financial appeal of their closed-weight counterparts.
Nonetheless, the model development strategies employed by Moonshot have sparked controversy and raised legal concerns. Earlier this week, Anthropic accused Moonshot of engaging in a prolonged campaign of model distillation, which allegedly involved directing nearly 300,000 requests from its Kimi model directly to Claude Opus, effectively substituting Opus responses for those generated by Kimi. Anthropic claimed that more than 23 million responses from its models were used in training Moonshot’s offerings.
