Moonshot AI launches Kimi K3 open-weight AI model with 2.8 trillion parameters

China-baseproximately 2.8 trillion parameters and a 1 million-token context window. The company says the model is designed for advanced reasoning, coding and knowledge-based tasks while supporting local deployment and customisation through its open-weight architecture.

China-based AI startup Moonshot AI has launched its new open-weight artificial intelligence model, Kimi K3. According to the company, the model has approximately 2.8 trillion parameters, making it the world's largest open-weight AI model.

Moonshot AI said Kimi K3 has approximately 2.8 trillion parameters and supports a 1 million-token context window. The company stated that the large context capacity enables the model to understand, analyse and respond to large volumes of information within a single prompt. According to the company, this capability increases its usefulness for research, software development, technical document analysis and other complex tasks.

According to Moonshot AI, one of Kimi K3's key features is its open-weight architecture. This allows developers and researchers to download the model weights, run the model on their own servers or computers, and customise it according to their requirements.

In contrast, the source states that most widely used AI models, including ChatGPT and Google Gemini, are primarily closed or proprietary models that are accessed through cloud platforms, with users not receiving direct access to the underlying model weights.

The company said open-weight models offer greater flexibility for research, enterprise AI solutions, data privacy and the development of customised AI applications for specialised industries. It added that demand for open-weight AI models has increased rapidly in recent years.

Moonshot AI launches Kimi K3 open-weight AI model with 2.8 trillion parameters

Moonshot AI said Kimi K3 was developed for advanced reasoning, long-duration coding projects and complex knowledge-based tasks. According to the company, the model has demonstrated high efficiency in technical areas such as GPU kernel optimisation, enabling better utilisation of AI hardware and lower latency.

The company also claimed that internal benchmark testing showed Kimi K3 performed at or above the level of several advanced AI models. However, broad independent verification of these claims has not yet been completed. The source states that experts believe the model's actual capabilities will become clear only after evaluation through public benchmarks and practical use.

The launch of Kimi K3 comes as several Chinese AI companies are expanding their presence in the global market. Besides Moonshot AI, companies including Z.ai and MiniMax are also developing large AI models at comparatively lower cost.

According to the source, this trend challenges the perception that cutting-edge AI development is limited to a small number of US companies. Chinese AI companies are seeking to strengthen their position in the global AI ecosystem through open-weight models, larger context windows and improved cost efficiency.

According to the source, if Kimi K3 performs in line with the company's claims, it could become an alternative for developers, startups and large enterprises. Organisations that prefer to run AI models on local servers instead of sending data to cloud platforms may benefit from open-weight models.

The source also states that research institutions and universities could use such models for AI research, language modelling, code generation and the development of industry-specific solutions.

 

Leave a comment