The launch of the Kimi K2 artificial intelligence model by Alibaba Group Holding-backed Moonshot AI has drawn rapid uptake amid praise from industry experts.
Downloads of Kimi K2, launched on July 11, doubled to 145,000 on Monday from 76,000 on Friday, according to AI and machine-learning developer platform Hugging Face.
The large language model (LLM) from the Beijing-based start-up uses a mixture-of-experts (MOE) architecture and boasts 1 trillion total parameters, with 32 billion activated per inference. For comparison, DeepSeek-V3 has 671 billion parameters.
“While companies like OpenAI invest hundreds of millions in compute resources, Moonshot’s Kimi K2 shows a more cost-efficient approach to training and inference – highlighting a possible turning point in AI development strategy,” said Henning Steier, Bluespace Ventures’ chief marketing officer, in a LinkedIn post.
The model is free via Kimi’s app and browser interface, unlike OpenAI’s GPT and Anthropic’s Claude, which charge monthly subscriptions.
MOE is a machine-learning approach that divides an AI model into separate sub-networks, or experts – each focused on a subset of the input data – to jointly perform a task. This is said to greatly reduce computation costs during training and achieve faster performance during inference.
Moonshot said it developed Kimi K2 at a fraction of the cost typically spent by larger AI firms.
