The trust gap in China’s AI models

Welcome to The Prompt, your Monday dive into the world of AI that puts Asia front and center. From Big Tech’s power players to the region’s scrappy disruptors, we cover it all. Want full access to all our AI reporting? Subscribe to us here.

---

Hello reader,

I recently went to a Muji store to grab some colored gel pens for my daughter. As I browsed through the selection, I noticed that some pens were labeled “Made in Japan,” while others had “Made in China” tags on them.

No offense to China, but since I know very well that Muji is a Japanese retailer, I picked the Japan-made ones.

Now I’m wondering if I tend to have the same thought process when choosing AI products. Do I lean toward large language models (LLMs) from Silicon Valley just because Google developed the foundational architecture behind them?

Speaking of AI, last week showed that the race between the US and China is far from over. Baidu released a new model with claims of better performance on vision-related benchmarks than OpenAI’s and Google’s, while Alibaba is working to revamp its mobile chatbot app to rival ChatGPT.

But the biggest one was Kimi K2 Thinking, a new model from Beijing-based Moonshot AI that launched to a great deal of fanfare.

Here comes another model that claims to outperform OpenAI’s and Anthropic’s in benchmarks while burning much less cash. CNBC reported that Kimi K2 Thinking only costed US$4.6 million to train, and Moonshot has since confirmed that the model was trained on Nvidia’s older H800 GPUs.

This allows Moonshot to set prices much more aggressively than its US competitors. When accessed through its API, Kimi K2 Thinking is 5x cheaper than Anthropic’s Claude Sonnet 4.5 for both input and output tokens.

MiniMax, another Chinese AI company, even went further and priced its recently released M2 model at 10x cheaper than Claude Sonnet 4.5. Yet despite all the efforts to make these advanced models cost-efficient, both Moonshot and Minimax haven’t been able to attract users.

Data from OpenRouter – a platform that provides access to many LLMs through API – shows that on November 12, users spent 94.7 billion tokens on Claude Sonnet 4.5, but only 6.3 billion on Kimi K2 Thinking and 38.8 billion on MiniMax M2.

But enough about the US-China showdown. Last week also gave us stories about how a startup called Prodigal built a profitable business out of an AI agent that collects debts with patience and empathy, and what it took for GitHub to hit record user growth in Singapore.

We also witnessed the quiet release of OpenAI’s GPT-5.1 and the impending departure of Meta’s chief AI scientist, Yann LeCun. It’s a lot to take in, I know, but that’s to be expected from a sector as hot as AI.

---

TOP AI READS FROM OUR DESK

---

WHAT WE THINK

---

STANDOUT AI DEALS

---