Standing up to Musk, where does the confidence of domestic large models come from?

There's been some significant news in the AI field these past two days. Elon Musk was asked how long it will take for China to catch up with Claude's top-tier Fable model in large models. Musk estimated it would be in the first quarter of next year. In the comments below, Tang Jie, the founder of Zhipu AI, replied that it might not take that long.

Let's look at the current ranking of domestic models. According to benchmark data from some ranking organizations, Zhipu AI's GLM5.2 is indeed second only to Claude's Fable in current benchmarks, and is even close to Opus4.8. From the perspective of the agent architecture, Zhipu's Zcode is also very close to and pays homage to Codex.

I've also tried Zhipu's models myself, mainly the Zcode agent. Overall, it does feel smoother than other domestic models. However, to say it can truly be compared to Codex and Claude, at least from my hands-on experience, there's still a gap. There are still many basic usability issues, like it freezing up when uploading multiple images and documents for analysis and learning simultaneously.

So my feeling is, if I can use the paid models of Codex and Claude daily, I would definitely prioritize those two. I might use Zhipu's Zcode agent as a backup.

I also shared an article on Planet about Morgan Stanley's focus points regarding China's AI development entering the 2.0 era, which will ultimately translate into specific areas to watch from an investment perspective. Those interested can take a look.

(Not for investment purposes)

$KNOWLEDGE ATLAS(02513.HK)  $Alibaba(BABA.US)  $Alphabet(GOOGL.US)  $Fullgoal SZSE ChiNext Artificial Intelligence ETF(159246.SZ)