Account subscription benefits alongside Premium Stories, Editorials,
Opinions and more. Unlock these with Subscription
The View From India
Looking at World Affairs from the Indian perspective.
First Day First Show
News and reviews from the world of cinema and streaming.
Today's Cache
Your download of the top 5 technology stories of the day.
Science For All
The weekly newsletter from science writers takes the jargon out of science and puts the fun in!
Data Point
Decoding the headlines with facts, figures, and numbers
THEdge
At the cutting edge of education and careers
Health Matters
Ramya Kannan writes to you on getting to good health, and staying there
Gender Agenda
Stories from beyond the binary.
The Hindu On Books
Books of the week, reviews, excerpts, new titles and features.
A Kimi logo is pictured at the Moonshot AI stand, featuring the Kimi K3 model, during the World Artificial Intelligence Conference in Shanghai on July 18, 2026.
| Photo Credit: AFP
Launch more U.S. open models, issue more U.S. visas, build more U.S. data centres, and encourage more U.S. competition — these were some of the panic-tinged solutions swirling online after the Chinese AI firm Moonshot released its Kimi K3 model on July 16.
Also Read | Is China’s AI bet turning American AI labs into the new Apple?
Kimi K3, with 2.8 trillion parameters and a 1 million-token context window, was billed as the “world’s first open 3T-class model”. By comparison, one of Google’s recent Gemma 4 open models reported a size of 30.7 billion total parameters. Moonshot AI candidly noted that the overall performance of Kimi K3 still “trails” Anthropic’s Claude Fable 5 and OpenAI’s GPT 5.6 Sol. However, the company claimed that Kimi K3 showed frontier-level performance across its evaluation suite. On July 27, the firm enjoyed a second burst of adulation when it released the model weights, allowing users to modify and run Kimi K3 as per their commercial needs under a far more flexible licence.
Kimi K3 cost about a third of Anthropic’s Fable model in terms of the input and output token prices. While business use cases, operational needs, and other overheads influence the final cost of deploying models, such price comparisons can put makers of more expensive models on edge.
