AI Research #Moonshot AI#Kimi K3#open weight#large language model#China#AI stocks#2.8 trillion parameters

Moonshot Launches Kimi K3: Largest Open-Weight AI Model at 2.8 Trillion Parameters

Moonshot AI released Kimi K3 as the largest openly available AI model with 2.8 trillion parameters. The model's open-weight release triggered a market sell-off in Chinese AI stocks, with Z.ai falling 30% and MiniMax dropping 16%.

Sunday July 26, 2026
Moonshot Launches Kimi K3: Largest Open-Weight AI Model at 2.8 Trillion Parameters

TL;DR

Moonshot AI released Kimi K3 as the largest openly available AI model with 2.8 trillion total parameters and 104 billion active parameters. The model’s open-weight release under a modified MIT license triggered a sharp sell-off in Chinese AI stocks, with Z.ai falling 30%, MiniMax dropping 16%, and Alibaba declining 4%. Moonshot’s daily revenue grew at least sixfold following the release.

The Model

Kimi K3 represents a significant leap in open-weight AI:

The model was released under a modified MIT license, making it one of the most permissively licensed large AI models available. Unlike many open-weight models that restrict commercial use, Kimi K3 allows unrestricted commercial deployment.

Benchmark Performance

Blind arena evaluations showed strong results:

The model’s coding performance is particularly notable. Independent testing showed K3 matching or exceeding the capabilities of much larger closed models on practical coding tasks.

Market Impact

The release triggered immediate market reactions:

The market reaction reflects concern that open-weight models could undermine the business models of companies selling API access to proprietary models. If K3 delivers comparable performance at a fraction of the cost, it could disrupt the revenue assumptions of several Chinese AI companies.

Industry Implications

Kimi K3’s release has several implications:

The release also raises questions about the future of AI business models. If open-weight models can match closed model performance, companies will need to differentiate on services, reliability, and specialized features rather than raw model capability.

Back to all news