- Moonshot AI unveiled Kimi K3 on July 16, boasting 2.7 trillion parameters, making it the largest open-weight large language model available today.
- Kimi K3 is said to perform competitively with Anthropic's Fable 5, the most advanced AI model widely available on the market today.
- Analysts were surprised by Kimi K3's capabilities, as they did not expect China to produce a model as powerful as Fable until early next year.
- Kimi K3 is priced at $15 per million output tokens, which is cheaper than Fable's $50 for the same amount, but more expensive than other Chinese models.
- Kimi K3's release has raised questions about the effectiveness of U.S. AI policy and the competitive landscape between U.S. and Chinese AI.
- The model's performance claims suggest that Chinese developers can build open-weight systems comparable to those from Anthropic and OpenAI.
- Kimi K3 has a 1 million-token context window, allowing it to process large amounts of text and work across text and images.
- In blind testing, developers preferred Kimi K3 over leading U.S. models, including Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol.
- Kimi K3's launch comes just before the 2026 World Artificial Intelligence Conference in Shanghai, where significant AI priorities are expected to be discussed.
- Moonshot AI raised $2 billion in funding in May, with a valuation exceeding $20 billion.
- The company is backed by major Chinese tech firms, including Alibaba and Meituan, as well as HSG (formerly Sequoia China).
- Moonshot AI's rapid advancements indicate that America's lead in advanced AI is diminishing as Chinese labs strive to make frontier-level intelligence more accessible.
Moonshot AI has unveiled the Kimi K3, a groundbreaking AI model featuring 2.8 trillion parameters, making it the largest open-weight system globally. This launch positions K3 as a formidable competitor to Anthropic's Fable 5, which is currently regarded as the most advanced AI model available.1
Analysts had anticipated that China would not produce a model of this caliber until early next year, marking K3's release as a significant milestone in Chinese AI development. If K3's performance claims are validated, it could indicate that Chinese developers are capable of creating open-weight systems that rival those of U.S. tech giants like OpenAI and Anthropic, potentially altering the landscape of global AI competition.6
K3 is priced at $15 per million output tokens, significantly lower than the $50 charged for Fable, raising questions about the sustainability of U.S. pricing strategies in the face of emerging Chinese competition. Yutong Zhang, president of Moonshot AI, emphasized the need for efficiency and fundamental research, stating, “We knew we didn’t have the luxury to simply scale up compute.”4
The model's capabilities include a 1 million-token context window, allowing it to process vast amounts of information simultaneously. Early benchmarks have shown K3 outperforming several leading U.S. models in specific tasks, further intensifying concerns in Silicon Valley regarding the competitive threat posed by Chinese AI advancements.7
As the AI landscape evolves, the launch of K3 could prompt a reevaluation of U.S. AI policies and strategies, especially with the upcoming 2026 World Artificial Intelligence Conference in Shanghai, where Chinese President Xi Jinping is expected to outline the nation’s AI priorities.9
“The 2.8 trillion-parameter model features a 1 million-token context window and outperforms leading US models in coding benchmarks, with developers preferring it over Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol. Mozilla CTO Raffi Krikorian said U.S. AI labs are 'clearly worried' as K3 costs $15 per million tokens, far less than Fable's $50.”
