The Chinese startup Moonshot AI launched Kimi K3 on Thursday, a language model with 2.8 trillion parameters, the largest open-source model to date. It surpasses GPT-5.5 and Claude Opus 4.8 in several areas, although it still trails behind the leaders GPT-5.6 Sol and Claude Fable 5.
The Chinese artificial intelligence startup Moonshot AI presented its large language model Kimi K3 on Thursday, July 16, which, with 2.8 trillion parameters, becomes the world's largest open-source AI model. The announcement shook the Asian tech market and reignited the debate over leadership in the global AI race.
Kimi K3: a quantum leap in parameters and capabilities
Kimi K3 is built on Moonshot AI's proprietary hybrid linear attention mechanism KDA and residual attention technology. The model natively supports visual understanding and has a context window of 1 million tokens, making it particularly suitable for long-range programming, intellectual work, and complex reasoning. The company will release the full model weights before July 27.
In internal evaluations, Kimi K3 has consistently outperformed GPT-5.5 from OpenAI and Claude Opus 4.8 from Anthropic in areas such as programming, visual understanding, and long-range task processing. However, the overall performance of the model still lags behind Claude Fable 5 and GPT-5.6 Sol, the most advanced models from its American rivals.
Immediate market impact and competition
Less than a day after its launch, Kimi K3 topped the front-end programming leaderboard on the global AI evaluation platform Arena AI, scoring 1.679, surpassing Claude Fable 5's 1.631 points and GPT-5.6 Sol's 1.618 points. This result demonstrates the model's competitiveness in specific tasks.
The impact on financial markets was immediate. Shares of Zhipu, listed in Hong Kong, plummeted by 28.49% to 1,107 HKD on July 17, while MiniMax fell by 15.62% to 216 HKD. BofA Securities noted that the new model shows that leadership in the AI industry is becoming ephemeral and that the 'open-source leader' narrative of Alibaba's Tongyi Qianwen is facing challenges.
Geopolitical context and sector outlook
The launch of Kimi K3 coincides with the opening of the 2026 World Artificial Intelligence Conference in Shanghai, featuring over 1,100 exhibiting companies and 3,000 samples. The performance gap between Chinese and American models continues to narrow, and the cost advantages of the former are attracting international market attention. At the same time, US lawmakers are considering restricting the adoption of Chinese AI models by domestic companies, adding uncertainty to the landscape.
For investors and industry professionals, the emergence of Kimi K3 serves as a reminder that competitive advantage in AI can be transient. Companies relying on proprietary models will need to closely monitor advancements in open-source, which offer increasingly powerful and cost-effective alternatives. Furthermore, the release of the full weights will allow developers worldwide to experiment and build on Kimi K3, accelerating innovation.
The next milestone will be the release of the model weights before July 27, at which point the AI community will be able to assess Kimi K3's capabilities firsthand. Until then, the market will remain attentive to the movements of OpenAI and Anthropic, which may respond with new versions of their own models.
You may also like
