Kimi K3 vs Qwen 3.8: the Chinese AI models taking on OpenAI

In the space of one week, two announcements out of China reshuffled the global AI landscape. Moonshot AI and Alibaba unveiled models that rival the best American ones — for a fraction of the price. Here's what that changes, and why it matters to you even 6,000 miles from Beijing.

Contents

1. The week that changed everything
2. Kimi K3 — The quiet giant
3. Qwen3.8 — Alibaba won't be left behind
4. Open-source vs proprietary: the real divide
5. The embargo paradox
6. What this means for a small business in the Basque Country
7. Frequently asked questions

1. The week that changed everything

Early July 2026. Within five days, two Chinese companies — Moonshot AI (Beijing, valued at $4 billion) and Alibaba ($200 billion market cap, Hangzhou) — dropped two bombshells on the global AI landscape.

Kimi K3 from Moonshot: 2.8 trillion parameters, open-weighted on July 27, ranked #1 worldwide on Arena.ai's Frontend Code Arena with a score of 1,679 — ahead of Claude Fable 5 (1,631) and GPT-5.6 Sol (1,618). A 17-spot jump in a single release.

Qwen3.8 from Alibaba: 2.4 trillion parameters, presented as "second only to Fable 5," open-weighted "soon."

"China's leading AI companies are ramping up the pressure on Silicon Valley — at a fraction of the cost."

The Verge, July 2026

This isn't the first time China has shaken up the market. DeepSeek V4 Pro (1.6T, MIT license) did it last year. Zhipu AI scored points with GLM-5.2 just last month. But this time is different: it's a double release, and both models are open.

2. Kimi K3 — The quiet giant

Moonshot AI isn't a newcomer. Founded in 2023 by Yang Zhilin (a Tsinghua PhD, ex-Google and Meta), the startup has been through some turbulence. Its Kimi assistant, once China's 3rd biggest AI app, dropped to 7th place after DeepSeek R1 launched in January 2025. Betting on open source has turned things around.

With Kimi K3, Moonshot is going big:

• 2.8 trillion parameters — 75% more than DeepSeek V4 Pro (1.6T)
• Architecture: sparse Mixture-of-Experts — 896 experts, 16 activated per token
• 1 million tokens of context
• Native vision (image and video) and a built-in reasoning mode
• Kimi Delta Attention (KDA): decoding up to 6.3x faster
• Pricing: $3/M input tokens, $15/M output
• OpenAI SDK-compatible API, open-source weights on July 27

3. Qwen3.8 — Alibaba won't be left behind

Alibaba didn't let Moonshot hog the spotlight. The very next weekend, the Chinese e-commerce giant unveiled Qwen3.8 (2.4T parameters), which it presents as "one of the most powerful models available today" and "just behind Fable 5."

With DeepSeek (MIT), Qwen (Apache 2.0), GLM 5.2 (Zhipu), and now Kimi K3, China is lining up a full arsenal of open models covering every segment: code, reasoning, multimodality, autonomous agents. This is an industrial strategy, not a one-off move.

4. Open-source vs proprietary: the real divide

The real divide isn't US vs China anymore. It's between open models and closed models.

OpenAI doesn't publish the weights for GPT-5.6 Sol. Anthropic keeps Claude Fable 5 under lock and key. The Chinese labs, on the other hand, put their best models in everyone's hands:

An important caveat: K3's hallucination rate rose from 39% to 51% between K2.6 and K3. It answers better but makes more up. Chinese benchmarks are self-reported — independent verification is still ongoing.

5. The embargo paradox: who's blocking whom?

A fact often overlooked: it isn't China that's closing itself off. It's the US government that banned selling high-end Nvidia chips to China. The embargo was meant to slow down Chinese AI by restricting access to GPUs.

The result: the opposite of what was intended. Cut off from the best GPUs, Chinese labs innovated on model architecture, training efficiency, and — above all — open source as a workaround strategy and a way to build global reach.

And Europe? No embargo on Chinese AI — but no champion approaching 2 trillion parameters either. France has Mistral, Germany has Aleph Alpha. Europe's gap remains an open question.

6. What this means for a small business in the Basque Country

Concretely, for a tradesperson in Anglet, an accountant in Bayonne, or a shop in Biarritz:

1. AI is getting cheaper. DeepSeek V4 Pro is 10x cheaper than GPT-5.6 Sol. GLM-5.2 is 5x cheaper. Chinese competition is putting huge pressure on prices — good news for small businesses.

2. Data sovereignty becomes achievable. An open-weight model hosted on a server in Europe means zero data leaving the country. For a small business handling customer data, that's a real, concrete GDPR lever.

3. The ecosystem is widening. It's not just ChatGPT anymore. There's Claude, Gemini, Copilot — and now Kimi, Qwen, GLM, DeepSeek. Every business can pick the tool that fits its trade, without depending on a single provider.

Of course, running a 2.8T-parameter model in-house remains out of reach without proper infrastructure. But distilled versions, rock-bottom API prices and shared hosting make these models accessible well beyond data centers.

❓ Frequently asked questions

Is Kimi K3 really better than Claude or ChatGPT?

Only on certain benchmarks. K3 ranks #1 on the Frontend Code Arena but its overall score is 57 versus 59-60 for the US leaders. On par with Claude Opus 4.8, but with a 51% hallucination rate.

Can I use Kimi K3 from the Basque Country?

Yes, via the Moonshot API (OpenAI SDK-compatible) or free chat at kimi.com. Open source since July 27 for self-hosting.

Do these Chinese models comply with GDPR?

Via API: data going to China = a GDPR problem. Self-hosted open source: no problem at all. You control where the model runs.

Which model should my small business choose?

GLM-5.2 for text (\$1.40/M tokens, MIT license). DeepSeek V4 Pro for rock-bottom pricing (\$0.44/M). Kimi K3 for code, if you can self-host it.

Why didn't the embargo slow China down?

Because the constraint forced innovation: more efficient architectures (MoE, linear attention), compression, and open source as a workaround strategy.

So, are you already using Chinese models in your daily work? Or does the "made in China" label still hold you back? I'd love to know what you think. Not as a techie — as a business owner, out of curiosity. Because this topic is too important to stay confined to insiders.