Chinese AI startup Moonshot has done something big! On Friday, the company launched Kimi K3. This is a huge 2.8-trillion-parameter model. Moonshot calls it the world’s largest open-weight AI system.
The company says it performs close to Anthropic’s frontier Fable model. That is an exciting claim in the fast-moving AI race.
A Fast-Narrowing Gap
The launch arrives at a striking moment. Just a month ago, the US government pulled Anthropic’s Fable and Mythos models over security concerns. Now China’s open AI ecosystem is catching up quickly.
Firms like Moonshot, Z.ai, and MiniMax keep releasing powerful models. Even better, they do it at much lower cost. This challenges the old Western belief that Chinese developers lag behind by months.
What Makes Kimi K3 Special
Kimi K3 is the first open-weight model to near the 3-trillion-parameter mark. It targets advanced reasoning, long coding tasks, and knowledge work.
Here’s a neat feature! The model offers a one-million-token context window. That means it can hold far more information in a single prompt than earlier versions.
Strong Benchmark Results
Moonshot shared some impressive numbers. Kimi K3 matched Fable 5 on GPU kernel optimisation. It also beat OpenAI’s Opus 4.8, GPT 5.6 Sol, and GPT 5.5. These techniques boost hardware use and cut delays.
Third-party testers agree. Arena.ai ranked Kimi K3 first for building web interfaces. Vals AI placed it second overall, behind Fable 5 but ahead of GPT-5.6 Sol.
Artificial Analysis found similar results. The model matched OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8 on complex, multi-step tasks.
Market Shake-Up
The news rattled competitors fast. Shares of Zhipu and MiniMax dropped sharply in Hong Kong. By midday, Zhipu fell 21.9 percent. MiniMax slipped 13.8 percent.
The Race Speeds Up
Chinese AI firms now release models faster than ever. Z.ai’s GLM-5.2 stunned observers by scoring near top US closed-source models.
MiniMax is not standing still. It plans a 2.7-trillion-parameter model as soon as the third quarter of 2026. A frontier multimodal model, H3, is also coming.
Earlier, Meituan’s LongCat-2.0 and DeepSeek’s V4-Pro led with 1.6 trillion parameters. Open-weight models let users download, run, and customise the systems. That sets them apart from closed models.
Smart Upgrades and Big Money
Moonshot added two key architectural upgrades. These improve efficiency and let the model handle long coding jobs with little human help.
The startup has strong backers, including Alibaba and Tencent. Bloomberg reported it seeks $2 billion in fresh funding. The valuation sits near $30 billion, ahead of a possible Hong Kong listing.
Read More: Pakistan to roll out AI diagnostic system in 1100 hospitals
