Kimi K3 and Qwen 3.8: Chinese AI shakes Silicon Valley again

When the Kimi K3 model had barely spent a week unsettling Silicon Valley, Qwen 3.8 arrived. They are two Chinese artificial intelligence models —open and highly advanced— presented at the World Artificial Intelligence Conference (WAIC) in Shanghai, and both revive an uncomfortable question for the big providers: why pay more for an expensive AI when a cheaper option already works well?

Not the first time

The scare already happened in January 2025 with DeepSeek: a Chinese company created a model that ranked well and was cheaper. A year and a half later, history repeats with K3, and not by chance: Chinese launches are often coordinated for WAIC's geopolitical showcase, and were partly created by distilling American models —a method forbidden by Western terms of service but practiced freely in China.

Are they really cheaper?

It depends on usage. As chatbots, Chinese models cost the same as ChatGPT or Claude: there is a free tier and paid plans from about 20 euros a month, with more generous limits on frontier models. The real difference is in the API: the per-token price is lower, which matters to companies and developers running the model hundreds or thousands of times programmatically.

Do they perform as well or better?

"It depends on the task, and that is the key," sums up Eric Risco, AI director at Andorra Telecom. In his experience, Kimi K3 is currently number one in frontend: he asked Kimi and Claude to build the same mini racing game with the exact same prompt, and Kimi's came out with better graphics and more detail. Developer Iván López, by contrast, finds Kimi K3 highly competitive in programming tasks for its cost-performance ratio, but Western models remain more consistent in aesthetic judgment and visual hierarchy.

There are also consistency nuances: "What Claude solves in one go, with Kimi you sometimes have to ask 5 or 10 times, correcting it, until it gets it right," warns Risco.

What it means for your company

The rivalry between providers is good news for users:

• Prices are falling. Chinese pressure forces everyone to cut costs, especially in automated API usage.
• Evaluate by task, not by brand. A model can be better at programming and worse at design; test with your real use cases before committing.
• Don't get locked in. Experts predict the mid-tier of models will converge: most will handle everyday tasks reasonably well and increasingly cheaply. Design your integrations so you can switch providers.

How we approach it at SimCodec

At SimCodec we help companies and SMEs choose and integrate artificial intelligence with sound judgment. With ten years of joint experience and EPIC as our regional strategic partner, we don't sell "the trendy AI": we assess your processes, compare models and APIs against your real use case, and build the integrations —custom software, cloud and automation— so technology delivers at the lowest possible cost. If you want to know whether Kimi K3, Qwen or any other model suits you, let's talk: we test it with your data and your tasks.