Moonshot launches 2.8-trillion-parameter Kimi K3, claims near-frontier AI performance

Kimi K3 incorporates two architectural upgrades that improve computing efficiency and enable it to complete long-horizon coding tasks with minimal human supervision.Chinese artificial intelligence (AI) startup Moonshot on Friday launched Kimi K3, a 2.8 trillion-parameter model that is said to be the world’s largest open-weight AI system and delivers performance approaching that of Anthropic’s frontier Fable model, Reuters reported.Moonshot said Kimi K3 is the first open-weight model to approach the 3-trillion-parameter mark and has been built for advanced reasoning, long-horizon coding, and knowledge-intensive tasks.

The model also features a one-million-token context window, enabling it to process substantially larger volumes of information in a single prompt than earlier generations.It added that Kimi K3 incorporates two architectural upgrades that improve computing efficiency and enable it to complete long-horizon coding tasks with minimal human supervision.According to the company, Kimi K3 “performed competitively with Fable 5 (with fallback) and substantially outperformed Opus 4.8, GPT 5.6 Sol, and GPT 5.5” in GPU kernel optimisation, a measure of how efficiently AI workloads utilise computing hardware while reducing latency.The company also cited third-party benchmark results, saying Arena.ai ranked Kimi K3 first in web interface-building capabilities, while Vals AI placed it second overall behind Fable 5 and ahead of GPT-5.6 Sol.

Artificial Analysis said the model delivered performance comparable to OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8, particularly on complex, multi-step reasoning tasks.Chinese companies, including Moonshot, Z.ai and MiniMax, have been releasing increasingly capable AI models at lower costs, challenging the perception that Chinese AI developers significantly lag their US counterparts.The launch comes a month after the US government withdrew Anthropic’s Fable and Mythos models over security concerns, highlighting the pace at which China’s open AI ecosystem is closing the gap with leading US developers.Chinese AI firms step up competitionThe momentum follows the release of Z.ai’s GLM-5.2, which achieved benchmark scores close to those of leading US closed-source models, challenging the view among many Western analysts that Chinese AI systems trail by at least six months.Hong Kong-listed MiniMax is also developing a 2.7 trillion-parameter model, expected to be released as early as the third quarter of 2026.

The company also plans to introduce its frontier-level multimodal model H3.Before Kimi K3, Meituan’s LongCat-2.0 and DeepSeek’s V4-Pro were among China’s largest AI models, each featuring 1.6 trillion parameters, while several other domestic developers have also crossed the trillion-parameter threshold.The push towards trillion-parameter models reflects increasing demand for AI systems capable of handling complex reasoning tasks.

AI developers are also working on systems capable of autonomous self-improvement, commonly referred to as recursive self-improvement.Unlike proprietary closed-source systems, open-weight models allow users to download, run and customise the underlying model.Backed by Alibaba and Tencent, Moonshot has been expanding its AI capabilities and raising capital to strengthen its position in the sector.

Bloomberg reported last month that the company was seeking $2 billion in fresh funding at a valuation of about $30 billion ahead of a potential Hong Kong listing.