In 1995, “The Legend of Sword and Fairy” burst onto the scene. Zhao Ling’er, Li Xiaoyao, and the ink-wash art style completely rocked the entire Chinese gaming community. Everyone was waiting for the sequel.
The wait ended up being eight years.
In January 2025, DeepSeek R1 was released. Trained at an extremely low cost, it produced a reasoning model capable of competing with OpenAI, causing NVIDIA’s stock to plummet and Silicon Valley to collectively lose sleep.
That feeling was exactly the same as when first booting up Pal all those years ago: so we can also make something like this.
Then everyone started waiting for R2.
Rumored for March, it was denied. Rumored for May, it didn’t arrive. Whispers said Liang Wenfeng was unsatisfied with the results, others said there was a chip shortage. 2025 ended, and R2 never came.
Around the 2026 Spring Festival, various new papers and code snippets emerged, raising expectations once again, yet R2 remained only in rumors.
Over a year has passed.
The lesson from waiting too long for Pal back then was profound: player expectations soared to the heavens, and when Pal 2 served up a half-finished product, its reputation crashed harder than if it had never been released. In contrast, Pal 3, released the same year, found a new breakthrough and salvaged everything.
Liang Wenfeng’s delay in releasing R2 might precisely be to avoid repeating Pal 2’s mistake. But the problem is, while Pal fans could wait eight years, the window of opportunity in the AI industry might only be a few months. OpenAI, Anthropic, Google, ByteDance, Alibaba—none of them are waiting for you.
It remains to be seen whether R2, when it arrives, will be a Pal 2 or a Pal 3.