Blab AI
Moonshot AI Targets $2 B Annual Revenue Amid Kimi K3 Usage Surge and Distillation Allegations
Moonshot AI, the Beijing‑based laboratory behind the world’s largest open‑weight model, has set its sights on $2 billion in annual revenue by the end of 2026—a leap that follows a sharp rise in its flagship model’s usage.
The ambition comes after Bloomberg reported a run‑rate of $1 billion in August, up from $300 million in June, and after the July release of Kimi K3. The company’s open‑weight strategy, which makes model weights freely available, is designed to drive high token volumes rather than premium pricing on the weights themselves.
Kimi K3 is a 2.8‑trillion‑parameter multimodal model that accepts images and video. It is the largest open‑weight model in existence and has been benchmarked against frontier models from OpenAI and Anthropic. The model’s architecture allows it to process longer sequences, and it has been adopted by developers for a range of applications, from chatbots to code generation.
OpenRouter data shows that K3 models generate roughly 300 billion tokens per day on the platform. At the current pricing—about $2.55 per million input tokens and $12.75 per million output tokens—this translates to roughly $4.6 million in daily revenue before revenue sharing. The custom license permits inference providers to share up to 30 % of revenue when they generate more than $20 million annually, a mechanism that keeps the model attractive to large-scale deployers while sharing the upside.
Moonshot’s target still falls short of the $40 billion and $65 billion annualized revenues reported for OpenAI and Anthropic, respectively. The open‑weight strategy lowers margins because the weights are freely available, but the high token volume suggests that monetization through API usage remains viable. If the token throughput remains steady, K3 would generate over $1.6 billion in revenue per year, underscoring the scale of the opportunity.
The aggressive plan has emerged amid a dispute with Anthropic, which accused Moonshot of running a large distillation campaign that routed nearly 300,000 requests from Kimi directly to its Claude Opus model. Anthropic alleged that Moonshot collected more than 23 million responses from Claude for use in training its own systems. The allegations were made public in a statement released earlier this week. Distillation—training a smaller or newer model on the outputs of a larger one—is common in the industry, but Anthropic’s claims imply that the requests were paid API calls, potentially violating the provider’s terms of service. No independent verification of the allegations has been reported.
Moonshot’s focus on open‑weight models reflects a broader trend in China’s AI ecosystem, where companies such as DeepSeek and Alibaba’s AI arm also release publicly licensed models. The high token throughput on OpenRouter indicates that developers are actively using K3 for a variety of applications, from chatbots to code generation. However, the company’s reliance on API revenue and the need for revenue sharing with inference providers may constrain profit margins.
As the AI industry continues to evolve, Moonshot’s $2 billion target underscores the commercial potential of open‑weight models, even as regulatory scrutiny and competitive pressures mount. The next steps will likely involve scaling infrastructure to support the projected token volume, addressing the distillation allegations, and navigating the regulatory environment in China and abroad. The outcome of these efforts will shape the future of open‑weight AI deployment and the broader competitive landscape.
The ambition comes after Bloomberg reported a run‑rate of $1 billion in August, up from $300 million in June, and after the July release of Kimi K3. The company’s open‑weight strategy, which makes model weights freely available, is designed to drive high token volumes rather than premium pricing on the weights themselves.
Kimi K3 is a 2.8‑trillion‑parameter multimodal model that accepts images and video. It is the largest open‑weight model in existence and has been benchmarked against frontier models from OpenAI and Anthropic. The model’s architecture allows it to process longer sequences, and it has been adopted by developers for a range of applications, from chatbots to code generation.
OpenRouter data shows that K3 models generate roughly 300 billion tokens per day on the platform. At the current pricing—about $2.55 per million input tokens and $12.75 per million output tokens—this translates to roughly $4.6 million in daily revenue before revenue sharing. The custom license permits inference providers to share up to 30 % of revenue when they generate more than $20 million annually, a mechanism that keeps the model attractive to large-scale deployers while sharing the upside.
Moonshot’s target still falls short of the $40 billion and $65 billion annualized revenues reported for OpenAI and Anthropic, respectively. The open‑weight strategy lowers margins because the weights are freely available, but the high token volume suggests that monetization through API usage remains viable. If the token throughput remains steady, K3 would generate over $1.6 billion in revenue per year, underscoring the scale of the opportunity.
The aggressive plan has emerged amid a dispute with Anthropic, which accused Moonshot of running a large distillation campaign that routed nearly 300,000 requests from Kimi directly to its Claude Opus model. Anthropic alleged that Moonshot collected more than 23 million responses from Claude for use in training its own systems. The allegations were made public in a statement released earlier this week. Distillation—training a smaller or newer model on the outputs of a larger one—is common in the industry, but Anthropic’s claims imply that the requests were paid API calls, potentially violating the provider’s terms of service. No independent verification of the allegations has been reported.
Moonshot’s focus on open‑weight models reflects a broader trend in China’s AI ecosystem, where companies such as DeepSeek and Alibaba’s AI arm also release publicly licensed models. The high token throughput on OpenRouter indicates that developers are actively using K3 for a variety of applications, from chatbots to code generation. However, the company’s reliance on API revenue and the need for revenue sharing with inference providers may constrain profit margins.
As the AI industry continues to evolve, Moonshot’s $2 billion target underscores the commercial potential of open‑weight models, even as regulatory scrutiny and competitive pressures mount. The next steps will likely involve scaling infrastructure to support the projected token volume, addressing the distillation allegations, and navigating the regulatory environment in China and abroad. The outcome of these efforts will shape the future of open‑weight AI deployment and the broader competitive landscape.
