That model, however, is now looking like the appetiser to Chinese AI firm Moonshot’s latest release, Kimi K3.

Kimi K3, which was officially launched late last week, is already looking to be making Moonshot a lot of money, with the company reportedly planning a public listing with a valuation well in excess of US$40 billion, and the model’s creator – 34-year-old ex Google hire Yang Zhilin – may be on to something.

The company has already been forced to pause sign-ups to the platform due to its initial popularity, and it also comes with a unique selling point, particularly for coders.

“Kimi K3 has strong long-horizon coding performance. Operating with minimal human oversight, it can sustain long engineering sessions, navigate massive repositories, and orchestrate terminal tools,” Moonshot said in a statement.

“Kimi K3 also excels in tasks blending software engineering with visual reasoning – it leverages screenshots and visuals to optimise game dev, frontend, and CAD.”

Moonshot contends that Kimi K3 can handle game development and visualisation, GPU compiling, and research coding. In addition, the mode can handle end-to-end knowledge work, video editing and animation, capable of creating entire short videos from several source clips and image files.

In terms of power, Kimi K3 isn’t the most powerful model out there, but nor is it lagging behind the rest of the pack. In fact, across a suite of benchmarks, the only other models to outperform Kimi K3 are Claude Fable 5 and GPT 5.6 Sol – and even then, it’s a near run thing.