According to CoinMeta, DeepSeek CEO Leung Man-fung revealed to investors that the company is training a model with 2 trillion parameters, and they also plan to develop an 8 trillion parameter model in the future. The current flagship v4-pro has only 1.6 trillion parameters. Based on public weights, moonshot ai kimi k3 has reached 2.8 trillion parameters, making it the largest open-source model in the world. The planned 8 trillion parameter model will be nearly three times that size. It's important to note that having more total parameters does not mean that all of them need to be used in each run; kimi k3 adopts the moe architecture, where only about 104 billion of the 2.8 trillion total parameters are activated at any given time.