Kimi K3 от Moonshot AI: эскалация гонки открытых весов и последствия для ИИ

Chinese lab Moonshot AI has released Kimi K3 — a new open-weight model, the successor to Kimi K2, built on a MoE architecture with roughly 1 trillion parameters. Researcher Nathan Lambert, writing on the Interconnects blog, calls the release an "escalation" of the open-weight race and breaks down its consequences for the entire global AI ecosystem.

What open-weight escalation means

Open-weight escalation is the practice of releasing models to the public that rival closed flagship models in capability. Kimi K3 continues this trend: its predecessor Kimi K2 launched in July 2025 as a MoE model with roughly 1 trillion parameters (about 32 billion active per token) with open weights.

Open weights mean anyone can download the model, run it on their own hardware, fine-tune it for their own task, and avoid paying for API access. This is exactly what sets the approach of Moonshot AI, DeepSeek, and Alibaba's Qwen apart from the closed OpenAI, Anthropic, and Google, which only provide access through paid interfaces.

• Developer — Moonshot AI (Beijing), the Kimi series
• Predecessor Kimi K2 — MoE with ~1 trillion parameters, ~32 billion active, open weights, July 2025
• Kimi K3 is positioned as an "escalation" of open weights
• Analyst behind the release — Nathan Lambert, Interconnects blog
• Today's leaders of the open-weights segment — China's DeepSeek, Qwen, Moonshot

Why this matters for the global ecosystem

Open weights out of China have already rattled markets once: in January 2025, the release of DeepSeek R1 triggered the largest single-day market-cap drop in history — Nvidia lost about $589 billion on January 27, 2025. Investors feared that cheap open models would undercut demand for expensive chips and subscriptions.

Kimi K3 reinforces that same dynamic. Every new powerful open-weight release narrows the gap between paid closed models and free open ones — and shifts the center of gravity of open AI toward China, while the West's Meta has visibly slowed its pace with the Llama lineup.

This is happening against the backdrop of US export controls that have restricted shipments of Nvidia's top-tier accelerators to China. Chinese labs are responding by betting on efficiency and openness: giving away strong models for free, building influence in places where they cannot buy an unlimited number of chips.

"Global consequences for the entire AI ecosystem" — that's how

Nathan Lambert sums up the essence of the Kimi K3 escalation on the Interconnects blog.

What it means

The open-weight race is turning into strategic rivalry. As labs like Moonshot AI release models at the level of Kimi K3 for free, the value of closed APIs falls, while the infrastructure and talent of open AI increasingly concentrate in China.

Frequently asked questions

What is Kimi K3?

Kimi K3 is a new open-weight model from the Chinese lab Moonshot AI, the successor to Kimi K2. Its weights are released publicly: the model can be downloaded and run independently, with no API fees.

Who is Nathan Lambert?

Nathan Lambert is a researcher focused on open models and the author of the Interconnects blog, where he analyzes the Kimi K3 release as an escalation of the open-weight race and its consequences for the global AI ecosystem.

• Meta has been designated an extremist organization and is banned in Russia.