Moonshot debuts Kimi K2 Thinking, Claims to surpass GPT-5 in agentic AI

Chinese startup Moonshot on Thursday unveiled its latest generative artificial intelligence (AI) model, “Kimi K2 Thinking,” which claims to surpass OpenAI’s ChatGPT 5 in “agentic” capabilities — or understanding what a user wants without explicit step-by-step instructions.

The model, backed by Alibaba, is built on the K2 model released in July by Beijing-based Moonshot.

Amid this update, Nvidia CEO Jensen Huang has once again called on the U.S. to accelerate its efforts in the competition with China over AI development. Simultaneously, some major U.S. firms, like Airbnb, have started publicly promoting certain Chinese AI models as equally viable, and often more cost-effective, alternatives to those offered by OpenAI.

Despite U.S. restrictions on Chinese businesses’ access to high-end chips, companies like DeepSeek have released open-source AI models that charge user fees significantly lower than those of ChatGPT. DeepSeek claims it invested $5.6 million in developing its V3 model, which is in stark contrast to the billions spent by OpenAI.

According to various tech sources familiar with the matter, the Kimi K2 Thinking model cost $4.6 million to train.

Moonshot claims that the Kimi 2 model can automatically select 200 to 300 tools to complete tasks on its own, lowering the need for human intervention.

The model processes up to 256,000 tokens in context length—roughly 200,000 words—and reportedly scored 87.3% on the Massive Multi-discipline Multimodal Understanding (MMMU) benchmark for multimodal reasoning, though these figures have not been independently verified. Moonshot plans to offer Kimi K2 through its existing chatbot interface and API, with tiered pricing that undercuts OpenAI’s rates by an estimated 30-40%.

Last month, DeepSeek unveiled a new AI model. This model reportedly enhances performance by utilizing visual cues to broaden the scope of the information it processes concurrently.