Chinese AI startup Moonshot has introduced Kimi K3, a 2.8 trillion-parameter open-weight model that the company claims is the largest system of its kind and delivers performance approaching Anthropic’s frontier Fable model, Reuters reports.

Moonshot Targets Advanced Reasoning and Coding

Moonshot describes Kimi K3 as the first open-weight model to approach three trillion parameters. It is designed for advanced reasoning, long-horizon coding, and knowledge-intensive work.

The model includes a one million-token context window, enabling it to process and retain significantly more information within a single prompt than previous generations.

Unlike proprietary models, an open-weight system allows users to download, operate, and customize its underlying model weights. However, Kimi K3’s scale means that relatively few users are expected to have the infrastructure needed to host it independently.

Moonshot said Kimi K3 “performed competitively with Fable 5 (with fallback) and substantially outperformed Anthropic’s Opus 4.8, GPT 5.6 Sol, and GPT 5.5” in GPU kernel optimization. The measurement assesses techniques used to maximize AI hardware utilization and reduce processing latency.

The company also said Kimi K3 incorporates two architectural improvements intended to increase computing efficiency and support long-horizon coding tasks with limited human supervision.

These performance and architecture claims come from Moonshot’s own technical disclosures rather than independent testing. The company’s full model weights are not yet public – they are expected to be released on July 27 – so some claims cannot be independently verified until then.

Independent Evaluations Strengthen Performance Claims

Third-party assessments placed Kimi K3 among leading models across several tests. Arena.ai ranked it first on a benchmark evaluating the ability to build web interfaces.

Vals AI ranked the model second overall, behind Fable 5 but ahead of GPT-5.6 Sol. Artificial Analysis found its performance comparable to OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8, particularly on evaluations involving complex, multi-step tasks.

The results suggest that China’s open-weight AI developers are narrowing the performance gap with leading proprietary systems from the United States.

Lower Costs Drive Chinese AI Adoption

Lian Jye Su, chief analyst at Omdia, said Chinese models are gaining traction because they can be deployed more cheaply than prominent U.S. systems.

“They can be run at a fraction of the cost that OpenAI charges its clients,” Su said. He cautioned, however, that a larger parameter count does not automatically produce better performance.

Why It Matters

Kimi K3 reflects intensifying competition between open-weight Chinese models and proprietary AI systems developed in the United States. The competitive focus is moving beyond model size to include reasoning, coding performance, infrastructure efficiency, and deployment cost.

For enterprises, open-weight models can provide greater control over customization and deployment. That flexibility must be balanced against the substantial computing resources required to host a system as large as Kimi K3.

The launch also demonstrates the faster release cycles emerging within China’s AI ecosystem. Backed by companies including Alibaba and Tencent, Moonshot has been expanding its technical capabilities and access to capital as it competes for a leading position in the sector.