Kimi K3: Has Chinese AI Finally Caught Up With America?

On July 16th, 2026, Moonshot released Kimi K3, an open-source 2.8T Mixture-of-Experts (MOE) model with a context window of up to 1 million tokens.

The Model Specs

The specs of Kimi K3 are crazy. It is an open-source 2.8T MOE model with 104B active parameters per token, native text, image, and video input, and a context window of up to 1 million tokens. Moonshot claims that Kimi K3 is the world's largest open-source AI model, and it is positioning it as a direct competitor to Anthropic's Claude 5 Opus, OpenAI's GPT 5.6 Sol, and Google's Gemini 3.1 Pro.

Benchmark Performance

Kimi K3 gets a score of 57 on the Artificial Analysis Intelligence Index, making it the #1 open-source model in the world right now, beating out GLM-5.2 and Google's Gemma 4 31B. It also sits squarely inside the top 5 models overall, right behind Claude Fable 5 (60) and GPT 5.6 Sol (59), and ahead of Grok 4.5 (54).

In coding, Kimi K3 scores an impressive 76.2% on Terminal-Bench 2.1 and 68.4% on SWE-bench Pro, matching or outperforming closed-source models that cost significantly more per request.

Why Chinese Open-Source Is Winning

While American labs have largely shifted toward proprietary, closed-door releases (with a few exceptions like Gemma and Inkling), Chinese labs have doubled down on open-weight architectures. By providing open access under permissive licenses, Chinese AI models are capturing developer mindshare worldwide.

Kimi K3 represents the pinnacle of this strategy: a model that matches American frontier performance while remaining open for the global developer ecosystem to inspect, self-host, and fine-tune.