Moonshot AI officially launched Kimi K3.1 Pro, the first model in the K3.1 series and its most capable 3-trillion-parameter model, supporting a 1M context window, with notable gains in coding, 3D creation, financial research, and other long-horizon agent capabilities. It also released Kimi K3.1 Pro Highspeed, offering the same intelligence at roughly 2-3x output speed.
SemiAnalysis reported that NVIDIA engineers optimized vLLM with large-scale megakernel fusion and Cute DSL-based decoding kernels, boosting the 2.8-trillion-parameter Kimi K3 model to 423 tok/s/user on GB300 NVL72 hardware, a 1.68x inference speedup.
Chen Guangyu, 17, announced on X that he had left Moonshot AI's Kimi team without disclosing reasons or next steps. He joined Kimi as an intern in November 2025 and co-authored the Attention Residuals paper with Zhang Yu and Su Jianlin, a technique adopted in Kimi K3 architecture.
At a DBS Bank forum, NVIDIA CEO Jensen Huang said Chinese AI models like DeepSeek, Kimi and GLM are already good enough. A firm open-source supporter, he said open source is the best way to improve efficiency, and the market will decide model use cases. He also predicted quantum computing could go mainstream in 3-5 years, rejected the idea of an end to AI competition, and argued cheaper, more open models will increase demand for Nvidia's chips, data centers and computing power.
Lip-Bu Tan said China's frontier AI models have low token costs and fast adoption, with DeepSeek, Kimi and GLM performing well enough. He firmly supports open source and believes the market will decide application scenarios for different models. He also said quantum computing may go mainstream in three to five years, and software will improve the efficiency of related infrastructure.
US Treasury Secretary Bessent claimed in an interview on October 3 that Chinese AI models conduct "industrial-scale distillation," calling it a euphemism for stealing US models. He also said one of China's stronger models, Kimi, believes itself to be Anthropic's Claude.
OpenAI announced on September 30 that it had encountered an organized adversarial distillation campaign starting in early July, and said the core group behind the activity is likely linked to Moonshot AI, developer of the Kimi model.
Baseten announced a partnership with OpenAI on September 30, becoming one of the first open-source model inference providers in OpenAI's B2B marketplace, allowing enterprise users to access Chinese open-source models such as Kimi K3 and GLM 5.3 via Codex or the Responses API.
On September 28, multiple developers discovered the "kimi-k3-1" identifier in the Moonshot AI API platform's backend model registry, which was detectable and callable via API probing. That evening, the Kimi official open platform showed a preview for K3.1 or K3 11111, supporting a 1M-token context window, with three reasoning intensity levels (Low, High, Max) expected, and possibly introducing Agent, Swarm multi-agent collaboration, search, and batch modes, with API pricing unchanged from the existing K3.
Kimi has launched a new browser extension (formerly Kimi WebBridge), allowing users to chat with Kimi in the browser sidebar to navigate websites, fill out forms, and record repetitive steps as reusable skills. The extension is now available on Kimi's official website and the Chrome Web Store.