After GLM-5.2, Kimi K3 signals China's AI race with the US is heating up

Only days after Chinese startup Z.ai drew attention with its GLM-5.2 model, another company has entered the spotlight. Beijing-based Moonshot AI has introduced Kimi K3, a massive open-weight model that the company says delivers performance approaching Anthropic's frontier Fable model across several demanding AI tasks.

Taken together, the launches point to a broader trend rather than an isolated success. Since DeepSeek's R1 surprised the industry last year, Chinese AI firms have continued to release more capable systems at a rapid pace, fuelling a growing debate over whether the gap with US developers has narrowed far more quickly than expected.

What is Kimi K3?

Moonshot describes Kimi K3 as the world's largest open-weight AI model, containing 2.8 trillion parameters. Unlike closed-source models, open-weight systems allow developers and organisations to download, customise and deploy the model themselves, making them particularly attractive for enterprises and researchers seeking greater control over AI deployments.

Beyond its sheer size, Kimi K3 has been designed to tackle demanding workloads involving complex reasoning, software development and knowledge-intensive tasks. The model supports a context window of one million tokens, enabling it to process and retain substantially larger amounts of information in a single prompt than earlier generations.

More from Tech

Moonshot has also introduced architectural improvements aimed at making the model more efficient while handling long-running coding tasks with minimal human intervention. According to the company, these upgrades improve computing efficiency without sacrificing performance.

The launch comes at a notable moment in the global AI landscape. Last month, Anthropic's Fable and Mythos models were withdrawn by the US government over security concerns, creating an opportunity for competitors to showcase alternative frontier-level systems.

How does Kimi K3 compare with leading AI models?

Moonshot claims Kimi K3 performs competitively with Anthropic's Fable 5 while surpassing several prominent AI models, including OpenAI's GPT-5.5, GPT-5.6 Sol and Opus 4.8, in GPU kernel optimisation, a measure of how effectively AI software utilises computing hardware while reducing latency.

The company also points to encouraging results from independent benchmarking organisations.

According to Arena.ai, Kimi K3 ranked first in tests evaluating web interface-building capabilities. Vals AI placed it second overall, behind only Fable 5 while ranking ahead of GPT-5.6 Sol. Artificial Analysis similarly found that the model delivers performance comparable to OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8, particularly on complex reasoning and multi-step problem-solving tasks.

While benchmark results should always be interpreted cautiously and may not perfectly reflect real-world performance, they nevertheless indicate that Chinese models are increasingly competing alongside the industry's best-known frontier systems rather than trailing them by a significant margin.

China's AI models are closing in on US rivals

The significance of Kimi K3 extends beyond a single product launch. It reflects a wider shift underway across China's AI ecosystem.

DeepSeek's R1 was the first major signal that Chinese developers could challenge assumptions about America's technological lead. Since then, companies including Z.ai, Moonshot and MiniMax have accelerated their release cycles, delivering increasingly sophisticated models while often keeping development costs lower than their Western counterparts.

Before Kimi K3, DeepSeek's V4-Pro and Meituan's LongCat-2.0 represented the country's largest publicly known models with around 1.6 trillion parameters. MiniMax is now reportedly preparing its own 2.7 trillion-parameter model alongside a frontier multimodal system known as H3.

The rapid pace of releases suggests Chinese developers are not relying on a single breakthrough but are steadily improving capabilities across reasoning, coding and autonomous task execution. Many of these models are also open-weight, helping expand adoption among developers and enterprises globally.

Moonshot itself has attracted significant backing from technology giants Alibaba and Tencent as it scales its AI ambitions. Bloomberg recently reported that the company is seeking around $2 billion in fresh funding at a valuation of roughly $30 billion ahead of a potential Hong Kong listing.

Whether Chinese companies have fully caught up with the very best American AI systems remains open to debate. Differences in real-world deployment, ecosystem maturity and product integration continue to matter. However, the latest wave of releases makes one thing increasingly clear: the conversation has moved beyond whether China can compete in frontier AI. The focus is now shifting to how quickly its companies can close whatever gap still remains.