China AI

AIArtificial Analysis reports that Nvidia's Vera Rubin serves Kimi K3 at about 2.8x the throughput per megawatt of GB300 at roughly 67 tok/s per user.

AIThird-party researchers known as the swarmchasers say agents running on Tencent Cloud made more than 2,600 attempts to extract information from Alibaba's Amap between September 28 and October 8.

AIStepFun says its Step 5 Preview is now live on DeepInfra, which hosts the model on day zero.

AIDeepSeek will refund all API users' spending from October 8 to 14 as credits after an outage on October 15 that disrupted its API from 8:11 to 13:04 Beijing time.

AIStepFun open-sourced Step 5 Preview, its new flagship model, on October 15, 2026.

AIA Chinese company's 8K AI spatial camera captures a Pixar campus scene, including a desk lamp, colored balls, trees and ground lighting, into a 3D world.

AITencent Hunyuan says recent reports of agent activity originated from evaluations of third-party models, not its Hy model.

AISome 95 per cent of Hong Kong investors want to outperform the market, the highest share in Asia where the average is 85 per cent, according to a August survey by Natixis Investment Managers.

AIStepFun thanks RadixArk for helping power Step 5 Preview in production with SGLang, which it says enables efficient, scalable inference for real-world workloads.

AIKindling AI releases an exlr8 K2/K3/K4 quantization of the 744B-parameter GLM-5.3 MoE model.

AIFal says Kling 4 is coming soon to fal. Kling AI's background post says Kling 4.0 is now fully available and shows a fantasy card example, "KING CARD," created with the model.

AIStepFun has released open weights for Step 5 Preview, a sparse MoE model with 600B total and 27B active parameters per token.

AIRadixArk links to a new SGLang cookbook covering the Step-5 Preview autoregressive model from StepFun. The post itself gives no further details on model size, benchmarks, or deployment steps beyond the link.

AIStepFun has released the BF16 and FP8 weights of Step 5 Preview on Hugging Face and ModelScope, according to RadixArk's post.

Why it matters: The post identifies the open-weight release channels and the serving stack, showing how a 600B-parameter MoE model is deployed through SGLang in production.

AITang Duyu, former head of Huawei's Xiaoyi multimodal large model, has founded Guanxing Shuzhi to build proactive physical AI systems that identify needs without explicit commands.

AIApple plans to launch Apple Intelligence in mainland China in December through the iOS 27.3 update, citing Bloomberg.

AIAnt Group's Ling-3.1-flash is an open-source model aimed at completing real workflows rather than leading benchmarks.

AISTEPX, a strategic partner of The Magic of Tomorrowland 2026, will bring its STEPX Neo agent phone to the Shanghai festival on October 17 and 18.

AIApple executive chairman Tim Cook said Apple Intelligence and the related Alibaba Qwen model have been approved and a beta version will launch later in October in China.

AIKilo says an early build of NVIDIA's Nemotron 3.5 Super VL scored 53% on KiloBench, the same as GLM 5.2.

AIArena says Xiaomi's MiMo V2.6 Pro ranks first among open models on Agent Arena with a +3.8% net improvement, after an update to its Bash Recovery signal.

AIDesign Arena announces that Pruna's P-Video-2-Pro-Reference is available on its platform.

AIMeta releases Muse Spark Open, an open-weight model with 816 billion parameters (42 billion active) and up to one million tokens of context.

AIVercel says US inference providers keep $2.50 of every $3.50 spent on Chinese open-weight models. The post gives no further figures or named providers.

AIDesign Arena says Pruna AI's P-Video-2-Pro-Reference is now available on its platform.

AINathan Lambert says US open-weight models were behind Chinese models dating to February 2026 until today. He says today's release cut that gap in time by at least half, to roughly three months.

AIVercel introduces Token Pass, which gives $80 in GLM 5.3 and Kimi K3 tokens for $30 per month.

AINick Cammarata, formerly of OpenAI's interpretability team and now at Goodfire, says neural networks are not tangled knots and that polysemanticity is wrong.

AIKling AI says Kling 4.0 is now fully available through a refreshed creation interface at kling.ai, plus MCP and CLI access for video creation workflows.

AISword Health built Dawn 1 by reinforcement learning post-training GLM-5.3 on prime-rl, Prime Intellect's framework.

AIOpenRouter says teams in its private beta are running dedicated GLM, Kimi, and DeepSeek deployments, plus reserved OpenAI and Anthropic capacity. The post links to a blog announcement on how to try the private endpoints.

AIMustafa Suleyman says Microsoft matched DeepSeek-V4-Pro-Base's pre-trained score on held-out code using 12x less training compute. He attributes the result to compounding efficiency gains.

AINvidia has released Nemotron 3.5 Super VL, a 120-billion-parameter open-weight model aimed at agentic work.

AIHuawei says its Next-Generation (NG) WAN Solution has entered large-scale commercial use and is deployed by more than 60 carriers worldwide, announced at UBBF 2026.

AIStepFun's open-weights Step 5 Preview, a 600B sparse MoE model with 27B active per token, is now served at FP8 on Featherless.

AIBlack Forest Labs says FLUX 3 Image Klein scores higher than Qwen Image 2.1, the current open-weights leader, in text-to-image and multi-reference generation in its internal evals.

AISemiAnalysis highlights a GLM-5 paper claiming full-stack adaptation to seven domestic Chinese chip platforms, including Huawei Ascend, Moore Threads, Hygon, Cambricon, Kunlunxin, MetaX and Enflame.

AIKling AI says Kling 4.0 is now fully available, accessible through a refreshed creation interface at kling.ai and through MCP and CLI for video creation workflows.

AIvLLM says DeepSeek-V4-Flash showed zero logprob difference at every step over 100 RL steps on 32 GB200 GPUs.

AIStepFun says its Step 5 Preview delivers strong inference performance on NVIDIA GB300 GPUs. NVIDIA says the 600B MoE model runs at nearly 8K tokens per second per GPU using NVFP4, and that it is StepFun's largest open-weights release.