China AI news, from the source
The model, price, policy and company changes that affect what you build outside China — verified against primary Chinese sources.
Save Events across devicesFollow companies and labsChina AI brief in your inbox
Top stories
Updated Oct 27, 2026 Share this briefingToday's brief is being compiled. Verified Events keep landing in the live update ledger below and in the full Event History.
Browse the Event HistoryMore China AI news
Evidence-qualified product, pricing, policy, capital and deployment Events, ordered by when they happened. Research appears only after an impact review; unverified Signals stay separate.
20 of 37 stories · last 7 days
Newest first. Wider windows expand what is available; open Event History for the full period.
Today · Oct 27, 2026 8 stories
Reportedly: Doubao Work adds an infinite creation canvas with Seedream 5.0 Flash and Doubao 2.1 Lite
ByteDance's Doubao Work has added an infinite creation canvas powered by the new Seedream 5.0 Flash image model and Doubao 2.1 Lite, a lighter model for everyday office tasks, Pan…
Why it matters ByteDance's Doubao Work users gain a new creation canvas and a lighter model for office tasks, potentially improving workflow efficiency.
Reportedly: AgiBot starts new funding round at about RMB 50 billion valuation
AgiBot is raising RMB 1–1.5 billion in a new round nearing completion at a valuation of about RMB 50 billion, IT Home reported, citing Caijing. The round means AgiBot cannot compl…
Why it matters AgiBot's new funding round delays its IPO until at least 120 days after the financing closes, which affects investors' exit timeline and may push the listing into 2027.
Reportedly: Moonshot AI teases Kimi K3.1 model in preview video
Moonshot AI released a 24-second preview video showing a computer running Kimi Code with 'Model: Kimi K3.1' displayed, IT Home reported. The video was published on October 23, 202…
Why it matters Developers using Moonshot AI's API will need to adapt to a new 'medium' thinking intensity option and potential changes in model routing when K3.1 becomes available.
inclusionAI released GLM-5.3-Flash-singprobe, a streaming safety probe for GLM-5.3-Flash
inclusionAI released GLM-5.3-Flash-singprobe, an intrinsic streaming guardrail built on zai-org/GLM-5.3-Flash, on Hugging Face. It adds less than 0.5% decode-time overhead and has…
Why it matters Developers using GLM-5.3-Flash can add streaming safety scoring with minimal overhead, avoiding a separate safety model.
inclusionAI released SingProbe, a streaming guardrail probe for Qwen3.8-Flash-Next
inclusionAI released SingProbe, an intrinsic streaming guardrail probe built on Qwen/Qwen3.8-Flash-Next, adding less than 0.5% decode-time overhead. It scores query intent, respon…
Why it matters Developers deploying Qwen3.8-Flash-Next can add streaming safety and hallucination scoring with minimal latency impact, avoiding a separate safety model.
Reportedly: DeepSeek open-sources operator toolchain with native Huawei Ascend support
DeepSeek released open-source versions of its operator development tool TileLang and five core compute/communication libraries (DeepGEMM, DeepEP, FlashMLA, TileKernels, DeepSelect…
Why it matters Developers using Nvidia's CUDA ecosystem now have an open-source alternative that runs on Huawei Ascend NPUs, potentially lowering the cost and complexity of porting AI workloads to Chinese-made…
Reportedly: Kuaishou Kling 4.0 video model fully launched, supports up to 30-second generations
Kuaishou's Kling AI announced the full launch of Kling 4.0 on klingai.com, with single video generations up to 30 seconds. 4K and 1080p 10-bit HDR output is coming soon, and API a…
Why it matters Kuaishou's full launch of Kling 4.0 puts pressure on competing video generation providers to match its 30-second duration and upcoming HDR output, potentially shifting enterprise expectations for…
Reportedly: Infinigence AI and Three-Body Universe launch Zhizi Agent 1.0, a multi-agent character agent
Infinigence AI and Three-Body Universe launched Zhizi Agent 1.0, a character agent based on the novel, QbitAI reported. It runs on Infinigence's AgentWorks platform with split inf…
Why it matters Developers outside China can evaluate Infinigence's AgentWorks platform and split inference approach for latency-sensitive character agents, but no API or open weights are confirmed in the evidence.
Yesterday · Oct 26, 2026 6 stories
Reportedly: Tencent Cloud open-sources TeamAI, a Git-based tool for sharing skills across agents
Tencent Cloud open-sourced TeamAI, an internal Git-based team AI collaboration tool that stores skills, work rules, tool configs, and project knowledge in a Git repo for review, u…
Why it matters Development teams using multiple AI coding agents gain a centralized way to manage and version shared skills and rules, reducing per-developer setup and inconsistency.
OpenBMB releases JustGoal GPU kernels outperforming expert libraries on NVIDIA B200
OpenBMB released the JustGoal repository with GPU kernels generated by a general-purpose agent given only an optimization goal. On NVIDIA B200, JustGoal outperforms open-source li…
Why it matters GPU kernel developers and AI infrastructure teams outside China can now access Apache 2.0-licensed kernels that outperform open-source libraries on NVIDIA B200, potentially reducing the cost of…
Reportedly: iFlytek releases Spark X2.6 end-cloud unified large model, first to achieve code and agent capability breakthrough on fully domestic compute
iFlytek released Spark X2.6, an end-cloud unified large model, at its 1024 Developer Festival, IT Home reported. The company claims it is the first to achieve code and agent capab…
Why it matters Chinese AI developers gain a model that reduces reliance on Nvidia GPUs, potentially lowering compute costs and easing export-control constraints.
Reportedly: Xiaomi open-sources MiDashengLM-Spatial, an end-to-end spatial audio understanding large model
Xiaomi released MiDashengLM-Spatial, an open-source end-to-end spatial audio understanding large model, with weights and inference code under Apache 2.0. It builds on MiDashengLM-…
Why it matters Developers outside China gain access to an open-source spatial audio model that can be integrated into smart home and automotive applications, potentially reducing the cost of building spatial audio…
inclusionAI released LemonDecider, a 4B decision model with explicit abstention
inclusionAI released LemonDecider, a 4B-parameter decision model derived from Qwen3.5-4B, on Hugging Face under Apache-2.0. It uses a 255-way single-token readout for choice, noul…
Why it matters Developers outside China can now use a specialized decision model from Ant Group's inclusionAI, reducing the need to adapt general-purpose LLMs for structured choice tasks.
Reportedly: Unitree Releases UniUni Swarm Control Software, Free for Its Users, Starting With H2 and G1
Unitree Robotics released UniUni, a free desktop swarm control app for Windows and macOS, initially supporting its H2 and G1 humanoids. Users can draw paths, arrange timelines, sy…
Why it matters Unitree's free swarm control software lowers the barrier for developers and researchers to coordinate multiple humanoid robots, potentially accelerating adoption in research and entertainment.
Oct 25, 2026 2 stories
BAAI open-sourced MS-Human-700, a 700-muscle human musculoskeletal model for reinforcement learning
BAAI released MS-Human-700, a whole-body human musculoskeletal model with 700 muscle–tendon units, on Hugging Face under Apache-2.0. The repository includes MuJoCo models, Gymnasi…
Why it matters Researchers and developers working on biomechanical simulation and reinforcement learning for humanoid control gain access to a detailed open-source musculoskeletal model and training code, reducing…
Reportedly: Alibaba Qwen open-sources BlenderWorld, a 223-task benchmark for 3D creation agents in Blender
Alibaba Qwen released BlenderWorld on October 14, an MIT-licensed benchmark with 223 tasks and 2,651 criteria for evaluating visual agents on real 3D creation in Blender. Code and…
Why it matters 3D designers and studios gain a standardized way to compare AI tools for Blender workflows, potentially reducing evaluation costs.
Oct 24, 2026 2 stories
Reportedly: Tencent WorkBuddy integrates Moonshot AI's Kimi-K3.1-Pro model with 1M context
Tencent WorkBuddy announced integration of Moonshot AI's Kimi-K3.1-Pro model, a 3-trillion-parameter model supporting 1 million context, IT Home reported. A high-speed variant off…
Why it matters Tencent WorkBuddy users gain access to a 3-trillion-parameter model with 1M context, potentially improving long-document and agentic workflows; competitors must match this integration to retain…
Reportedly: Unitree Robotics releases H2-X Kung Fu Legend Guardian humanoid robot that can lift a car and kick in a car door
Unitree Robotics released the H2-X Kung Fu Legend Guardian humanoid robot on October 21, 2026, with up to 8 hours of battery life. It can lift a car with bare hands and kick in a …
Why it matters Unitree Robotics' demonstration of a humanoid robot performing high-force tasks could pressure competitors to disclose comparable physical capabilities, affecting procurement evaluations for…
Oct 23, 2026 2 stories
Reportedly: MiniMax releases M3.1 Flash: 3x faster inference, 70% lower cost
MiniMax released the M3.1 Flash model, previously anonymous as Space Bunny, with nearly 3x faster inference and 70% lower cost. It scores 76.2 on BankerToolBench and 73.7% on Deep…
Why it matters Developers using MiniMax's API can now access faster and cheaper inference for agentic and coding workloads, potentially reducing their operational costs.
Reportedly: ByteDance Volcano Engine releases Seedance 2.5 4K model for cinematic-quality video generation
ByteDance's Volcano Engine released Seedance 2.5 4K, the strongest model in the Seedance 2.5 generation, trained specifically on facial expressions, subtle movements, and image de…
Why it matters Film and television production companies gain a cost-efficient workflow: they can iterate on 480P drafts before committing to expensive 4K renders, reducing production costs and time.
Reportedly: Alibaba's Qoder Adds Loongson LoongArch Support Alongside UOS and Kylin Desktops
Alibaba's Qoder coding agent now runs on Loongson LoongArch64 as well as Linux ARM64, with builds for UnionTech UOS Desktop V20 and Galaxy Kylin Desktop V10, Pandaily reported.
Why it matters Developers on Chinese domestic CPU architectures and operating systems gain access to Alibaba's Qoder coding agent, expanding the tool's reach in China's self-reliance ecosystem.
Reportedly: Shanghai AI Laboratory Releases Atria Dawn Full Version With a 512K Context Window
Shanghai AI Laboratory released the full version of its Atria Dawn agent model, built on a 744B MoE GLM-5.2 base, with a 512K context window. The release improves scores on JobBen…
Why it matters Developers outside China cannot yet access the weights, so they cannot evaluate or deploy the model; the 'Coming soon' status means the open-weights release is still pending.
Reportedly: OpenCode makes DeepSeek V4.1 Flash free with 200K context
OpenCode announced DeepSeek V4.1 Flash is now free, with the free tier supporting a 200K context window, down from the 1M version, IT Home reported. The move follows earlier quota…
Why it matters Developers using OpenCode can now access DeepSeek V4.1 Flash without quota limits, lowering their cost for tasks within 200K context.
Oct 22, 2026 0 stories
Reportedly: MiniMax shares fall below IPO price after 7x surge
MiniMax's stock fell as much as 7.3% to HK$159.3 on October 22, below its HK$165 IPO price, IT Home reported. The company listed on January 9, 2026, raising about HK$5.54 billion.…
Why it matters MiniMax's share price decline shows that investors are reassessing the value of Chinese large model companies amid intensifying price competition and persistent losses.
No additional verified events in this window. Try a wider period or open Event History.
View All eventsNew Chinese AI models
All models- XiaomiMiMo-V2.6-ProSep 21, 2026 · Open weights
- DeepSeekDeepSeek-V4.1-FlashSep 10, 2026 · Open weights · Global API
- Zhipu AIGLM-5.3-FlashAug 26, 2026 · Open weights · Global API
- AlibabaQwen3.8-Flash-NextAug 26, 2026 · Open weights · Global API
- Zhipu AIGLM-5.3Aug 18, 2026 · Open weights · Global API
- AlibabaQwen3.8-27BAug 14, 2026 · Open weights · Global API
Explore China AI
- China AI modelsAccess, open weights, and licenses across Chinese labs.
- Chinese AI companies73 frontier labs, big tech, robotics and AI chip makers.
- China AI policyRules, algorithm filings, labeling standards and export controls.
- Attention IndexEnglish and Chinese source attention.
- China AI ExplainersEssential primers on terminology, frontier startups, and compliance.
- China AI LinesFive strategic themes: open weights, price wars, robotics, chips and policy.
Your reading list
Save stories. Free, across devices.