What is Kimi K3? Why this new Chinese AI model is making headlines

China’s AI startup company, Moonshot AI, released a new version of its Kimi AI model called Kimi K3. The company claims that its “world’s largest open-weight AI model” that it challenges frontier US AI models like Anthropic’s Claude Fable 5. 

Moonshot AI also shared benchmark scores, claiming that Kimi K3 offers improved performance compared to AI models like OpenAI’s GPT-5.5, GPT-5.6 Sol and Opus 4.8. The company, in a press statement, said, “K3 stands as Moonshot AI’s most powerful open-source coding model to date.” Here’s what you need to know about this new Chinise AI model.

Must read: 'It doesn't make sense': Satya Nadella criticises Anthropic's Fable 5 control and restrictions

What is Kimi K3?

Kimi K3 is the new-generation AI model by Moonshot AI, which has 2.7 trillion parameters, offering the ability to learn complex patterns and handle advanced tasks. For comparison, DeepSeek V4 has 1.6 trillion parameters, meaning Kimi K3 has a significantly larger parameter count. In addition, it supports a context window of one million tokens.

Kimi K3 is an open-weight model, giving enterprises and researchers the ability to download, fine-tune and deploy the model independently. This offers greater control over data, infrastructure and AI development. Moonshot AI claims that Kimi K3 is built for complex and demanding tasks that involve reasoning, software development and knowledge-based tasks.

Must read: What is 'LOL' access? 3 trade secrets Apple says a former engineer took to OpenAI

The company said Kimi K3 can operate with “minimal human oversight, it can sustain long engineering sessions, navigate massive repositories, and orchestrate terminal tools.”

Kimi K3 performance compared to leading AI models

Moonshot AI claims that Kimi K3 delivers performance similar to Anthropic's Claude Fable 5, which is known as one of the most advanced AI models currently available. In addition, it says the model “substantially outperformed” Anthropic's Opus 4.8 and OpenAI's GPT 5.6 Sol and GPT 5.5.

Moonshot's benchmark results suggest Kimi K3's biggest strengths lie in agentic coding, software engineering and complex reasoning. Kimi K3 outperformed Anthropic's Claude Fable 5 on BrowseComp, Program Bench and SWE Marathon. Against OpenAI's GPT-5.6 Sol, Kimi K3 scored higher on FrontierSWE, SWE Marathon and DeepSearchQA.

This showcases that China’s AI ambitions are closing the gap with leading US AI companies, with its latest models competing neck-to-neck with top US-developed AI models across key performance benchmarks.