Goldman Sachs comments on Kimi K3: China's open-source models have reached a global tipping point in "intelligence," with fierce competition among high-end models.

```

The launch of Kimi K3 marks a new stage for China’s open-source AI models in the global competitive landscape—model intelligence has reached the critical point for worldwide mass adoption, pricing power is rising, but fierce competition in the high-end programming sector is putting pressure on related stocks.

According to Wind Chasing Trading Desk, Goldman Sachs pointed out in a research report released on July 18th that Kimi K3 was officially launched on July 17th, with a parameter scale of 2.8 trillion, and ranks at the forefront globally in both Arena.ai programming rankings and Artificial Analysis intelligence scores. At the same time, Kimi K3 set its mixed API pricing at $2.3 per million tokens, setting a new high for model pricing in China, demonstrating that Chinese AI model companies are gradually transitioning from “cost efficiency” competition to “pricing power.”

On the day Kimi K3 was released, market concerns about China’s AI model competitive landscape were quickly reflected in stock prices. Zhilu fell 28% on the day, MiniMax dropped 16%, showing investors’ high uncertainty regarding the long-term competitiveness and sustainability of leadership among Chinese AI model companies.

Kimi K3 Sets a New Benchmark for Model Pricing in China

Kimi K3 was launched with 2.8 trillion parameters, achieving global-leading levels in programming capabilities and some agentic features. The report cites Arena.ai programming rankings and the Artificial Analysis intelligence index, showing that K3’s overall performance is close to or on par with the world’s most advanced models (SOTA).

In terms of pricing, K3 set mixed API pricing at $2.3 per million tokens, significantly higher than comparable domestic products: Alibaba’s Qwen3.7 Max is $1.4, Zhilu GLM5.2 is $0.9, MiniMax M3 is $0.22, DeepSeek V4 Pro is $0.18. Although K3’s pricing is still lower than global SOTA models, its pricing strategy clearly shows the path to greater pricing power as model intelligence increases.

Goldman Sachs believes that the intelligence of China’s open-source/open-weight AI models has reached a critical node for global mass adoption, consistent with the core perspective of its previously released China LLM (Large Language Model) research report.

High-End Programming Track Continues to Heat Up

After Kimi K3’s launch, more Chinese AI models will be intensively released in the second half of the year, including Zhilu GLM, Alibaba Qwen, MiniMax M3 Pro, etc., with parameter scales further expanding to the 2-5 trillion range. Competition in the programming/code generation track will remain fierce, and integration of multimodal and visual understanding abilities will become the next major upgrade direction for China’s foundational models.

Each model company is actively laying out agentic applications as key traffic entry points, especially in programming fields, such as Zhilu’s ZCode, Tencent’s Workbuddy, Alibaba’s Qoder and other aggregation platforms, aiming to obtain more real programming and agentic data in a closed loop to support model iteration. Collaborative office and industry expert agent products may become the next priority direction.

In the low-end pricing sector, API pricing and gross margin for agent applications are expected to remain under pressure, with price competition fierce in the $0.1 to $0.2 per million tokens range. The reason is that, following financing, Chinese AI model companies generally possess ample cash reserves, enabling them to subsidize pricing. Therefore, financial strength is listed as one of the three core indicators for evaluating Chinese AI model companies, alongside pricing power and cost efficiency.

In contrast to the price war in the text foundational model market, Goldman Sachs holds a more optimistic view for video generation models. The report notes that SeeDance, Kuaishou, as well as MiniMax’s Hailuo and the upcoming H3 model, are expected to maintain healthy growth in the latter half of 2026, benefiting from breakthroughs such as the integration of video generation and large language models, persistent supply constraints in computing power, and demand significantly exceeding capacity.

~~~~~~~~~~~~~~~~~~~~~~~~

The above exciting content is from Wind Chasing Trading Desk.

For more detailed interpretations, including real-time analysis and frontline research, please join 【Wind Chasing Trading Desk ▪ Annual Membership】

Risk Warning and DisclaimerThe market has risks, investment requires caution. This article does not constitute personal investment advice and does not take into account individual users’ specific investment objectives, financial situation or needs. Users should consider whether any opinions, views or conclusions in this article suit their specific situation. Investing accordingly is at your own risk. ```