March 27th. March 4thLin Jun-hyunSCRAMBLES IN X, BID FAREWELL TO THE LARGE OPEN-SOURCE MODEL PROJECT。
In the early hours of March 4th, the Qwen technical director, Lin Jun-chun, suddenly wrote a farewell to his large open source model project。
Almost a month after leaving Ali, Lin Joon-soo published a long article on social networking platform, stating that the first wave of the reasoning model, represented by OpenAI o1 and DeepSeek-R1, validated the value of enhanced learning in the post-training phase, but that the core of the next phase of the industry would turn to "smart body thinking" , i.e. the model's continuous revision of strategies through interaction with the real environment and in action。
In addition, Lin Jun-chun has a rare reset of Qwen3 early attempts to combine the limitations of the "think" and "directive" models。
He admitted that forced integration could easily lead to a compromise in model performance because of the structural contradiction between the command scene for efficient implementation and the considered reasoning scene at the bottom of the data requirements。
This explains Qwen's re-diversification of the Instract and Thinking stand-alone product lines in subsequent editions (e.g. version 2507) to meet the high-volume and low-cost demand of commercial customers。