Moonshot AI's Kimi K3 rivals Anthropic Fable 5 with 2.7T parameters

Chinese AI model reaches frontier performance tier

Chinese startup Moonshot AI released Kimi K3 on July 16, an open-weight large language model with 2.7 trillion parameters that the company claims performs competitively with Anthropic's Fable 5, currently the most advanced AI model in wide commercial use. The release narrows the performance gap between Chinese and U.S. AI systems significantly earlier than industry analysts anticipated.

Kimi K3 represents the largest open-weight LLM available today, exceeding DeepSeek V4's 1.6 trillion parameters. According to Moonshot's benchmarks, K3 consistently ranks among the top three models and substantially outperforms Anthropic's Opus 4.8 and OpenAI's GPT 5.6 Sol and GPT 5.5.

Why it matters

The timing and capability of K3 challenges assumptions about how effectively U.S. export controls can slow Chinese AI development. Analysts had not expected China to produce a Fable-class model until early 2027. The release arrives as global enterprises increasingly question the cost of deploying models from Anthropic and OpenAI, with K3 priced at $15 per million output tokens compared to Fable's $50. This cost advantage, combined with the model's open-weight architecture, could accelerate adoption among developers seeking alternatives to U.S.-based closed systems.

U.S. policy implications

The development could reshape U.S. AI policy in competing directions. The U.S. government previously imposed temporary export controls on Anthropic's Mythos and Fable models after researchers demonstrated ways to bypass safety guardrails. Officials also initially restricted OpenAI's GPT-5.6 release to select partners.

U.S. politicians are now considering measures to prevent Chinese developers from using "distillation" techniques—training smaller models using outputs from larger, more powerful systems. Anthropic has accused Moonshot, along with z.ai, Minimax, Alibaba, and DeepSeek, of conducting illicit distillation attacks. Policymakers are also exploring ways to counter the appeal of Chinese open-source models, potentially by encouraging domestic open-source alternatives.

Technical efficiency under constraints

U.S. export controls barred Chinese developers from accessing the advanced AI processors used to train cutting-edge models, forcing them to optimize for computational efficiency. Yutong Zhang, president of Moonshot AI, acknowledged this constraint at the World Economic Forum earlier this year: "We knew we didn't have the luxury to simply scale up compute. That forced us to focus on fundamental research and efficiency."

Moonshot's earlier models had already gained traction in Silicon Valley. Cursor, a coding startup, used Kimi to build Composer 2, its AI coding agent. DoorDash chief technology officer Andy Fang noted in early July that the company delegates lower-level work to Kimi K2.6. Thinking Machines used Kimi K2.5 to generate post-training data for its Inkling model, released July 15.

Commercial trajectory

Moonshot AI raised $2 billion in May at a valuation exceeding $20 billion, with annual recurring revenue surpassing $200 million. The company's backers include Alibaba, Tencent, Meituan, and Hongshan Capital. Moonshot is reportedly preparing for an initial public offering in Hong Kong, following fellow Chinese AI developers MiniMax and z.ai, which went public in Hong Kong in January.

These details were first reported by Fortune.