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 Nov 16, 2026 Share this briefing
01of 3

Reportedly: Tencent releases Hy4 flash: 224B total parameters, 12B active, per-task cost about 5% of Hy4

Read story and sources Share

What happened

Tencent's Hunyuan team released Hy4 flash, a 224B-parameter model with 12B active per token, 1M-token context, and multimodal input. API pricing is much lower than Hy4; on WorkIndex, per-task cost is 1/20 of Hy4 and 1/8 of DeepSeek-V4.1-Flash-1017, IT Home reported.

Why it matters

Developers using Tencent Hunyuan APIs can now get Hy4-level intelligence at about 5% of the per-task cost, directly undercutting DeepSeek's flash-tier pricing by 8x on the cited benchmark. This pressures rival Chinese model providers to match the price-performance or lose cost-sensitive enterprise …

What to watch next

Watch for independent verification of the WorkIndex cost claims and real-world latency at 1M context. A key signal is whether Tencent extends the same Engram/LatentMoE architecture to larger Hy models or open-weights releases.

02of 3

DeepSeek Releases Full-Suite App API

Read story and sources Share

What happened

DeepSeek released a Full-Suite App API. The source announcement was published on November 13, 2026. Availability details are not specified in the evidence.

Why it matters

Developers outside China may gain a new integration point for DeepSeek's applications, but the evidence does not confirm whether the API is accessible globally or limited to China. The specific impact on cost, access, or competitor positioning cannot be determined from the available information.

What to watch next

The next verifiable signal is the publication of detailed API documentation or a changelog specifying endpoints, authentication, rate limits, and regional availability. Monitor the DeepSeek API News page for follow-up announcements.

03of 3
Open weights

MiniMaxAI open-sourced MiniMax-H3-Image-PE, a prompt expansion model for MiniMax-H3-Image

Read story and sources Share

What happened

MiniMaxAI released MiniMax-H3-Image-PE on Hugging Face, a prompt expansion model that turns short requests into detailed image descriptions and aspect-ratio recommendations for MiniMax-H3-Image. The model is fine-tuned from Qwen3.8-27B in FP4 and released under the MiniMax-H3-Image Community License Agreement.

Why it matters

Developers outside China can now use MiniMax-H3-Image-PE to improve image generation prompts for MiniMax-H3-Image, potentially reducing the need for manual prompt engineering. The release of prompt expansion weights under a community license may lower the barrier to integrating MiniMax's image gene…

What to watch next

The image editing prompt expansion is still under development, with only the prompt released. A specific signal to watch is whether MiniMaxAI releases the image editing PE weights and whether the model gains adoption in image generation workflows outside China.

AIGC Newsletter

DeepSeek DeepInfer, open-sourced: the engine behind 155 trillion tokens a day — and the unit economics DeepSeek just published with it

Three days after open-sourcing the model, DeepSeek open-sourced the machine that serves it: a production inference engine under Apache 2.0, serving economics no other lab publishes — 155 trillion tokens a day from one cluster, US$38 revenue per card-hour — and a hardware list that names NVIDIA and Huawei in the same breath.

More 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 27 stories · last 7 days

Newest first. Wider windows expand what is available; open Event History for the full period.

Yesterday · Nov 16, 2026 3 stories

NEW · Moonshot AI · Reported by Leiphone

Reportedly: Qianwen Office adds 10 models including Qwen4-Max, Kimi-K3.1, GLM-5.3

Qianwen Office added 10 models on Nov 16, including Qwen4-Max, Kimi-K3.1, GLM-5.3, DeepSeek, MiniMax, and Step. Personal and enterprise standard users can now choose models freely…

Why it matters Qianwen Office users gain lower token costs and better task fit by choosing among 10 models, while enterprise admins get granular control over model access.

NEW · Huawei · Reported by IT Home

Reportedly: Huawei Kirin 9050 series chip details revealed: 30% audio performance boost, first H.266 hard decoding, low-latency AI heterogeneous computing

Huawei revealed new details of its Kirin 9050 series chip, including a 30% audio performance boost, first H.266 hard decoding, and low-latency AI heterogeneous computing. The chip…

Why it matters Huawei's Kirin 9050 series with H.266 hard decoding support could pressure other chipmakers to adopt the newer codec, potentially reducing bandwidth costs for streaming services and improving user…

NEW · Alibaba · Reported by IT Home and SCMP Tech

Reportedly: Alibaba releases Qwen4-Max, a 2.4-trillion-parameter MoE flagship model with 1M-token context and over 60% lower task cost

Alibaba's Qwen team released Qwen4-Max, a 2.4-trillion-parameter sparse MoE flagship model with native vision and a 1M-token context window. It is available via the Qwen AI platfo…

Why it matters Developers using Qwen's API face lower per-task costs and a longer context window, which may reduce the need for external retrieval or summarization pipelines.

Nov 15, 2026 2 stories

NEW · Huawei · Reported by IT Home

Reportedly: Huawei Cloud launches open-source Pangu model openPangu-2.1-Pro on MaaS platform

Huawei Cloud released openPangu-2.1-Pro on its MaaS platform, upgrading agent development, AI coding, and inference optimization. The model is optimized with over 100,000 data ent…

Why it matters Developers using Huawei Cloud gain access to a model optimized for HarmonyOS and Ascend, potentially reducing development effort for those ecosystems.

NEW · Infinigence AI · Reported by Pandaily

Reportedly: D-Robotics and Infinigence AI Bring RLinf and Three Other Open-Source Robot Tools to Sunrise Chips

D-Robotics and Infinigence AI linked RLark, RLinf, RPent and APXInf to Sunrise chips and RoboGo. RLinf is on RoboGo, and APXInf runs Pi0.5 on Sunrise S600, Pandaily reported.

Why it matters Robot developers using Sunrise chips gain access to open-source RL and inference tools, reducing the cost and complexity of building embodied AI applications on D-Robotics hardware.

Nov 14, 2026 2 stories

NEW · Alibaba · Official source

Qwen releases Qwen-Live-Delivery-Plugin for human-in-the-loop long-running AI agents

Qwen (Alibaba) released Qwen-Live-Delivery-Plugin 0.1.0, an Apache-2.0 licensed TypeScript plugin for Codex CLI that lets long-running AI agents deliver iterative versions while u…

Why it matters Developers using Codex CLI gain a way to supervise long-running agent tasks without losing delivered work, reducing the risk of wasted compute and unusable outputs.

NEW · Moore Threads · Reported by IT Home

Reportedly: Tsinghua University releases ProtCell single-cell proteomics foundation model, reproduces pretraining on Moore Threads MTT S5000

Tsinghua University's Jiang Rui group released ProtCell, a single-cell proteomics foundation model pretrained on ~380 million cells from 98 public datasets. The team reproduced fu…

Why it matters Chinese life-science AI teams gain a domestic GPU option for large-scale pretraining, reducing dependence on restricted imported accelerators for workloads involving sensitive genomic and proteomic…

Nov 13, 2026 4 stories

NEW · Tencent · Reported by IT Home and Pandaily

Reportedly: Tencent Hunyuan releases real-time speech model Hy Realtime with tool calling and singing, live in Yuanbao voice calls and Tencent customer service

Tencent Hunyuan released Hy Realtime, a real-time speech model with agentic tool calling and singing, on November 13, 2026. It is live in Yuanbao voice calls and Tencent customer …

Why it matters Tencent's integration of Hy Realtime into Yuanbao and customer service gives it a direct distribution channel to millions of users, potentially reducing reliance on third-party speech models and…

NEW · Alibaba · Reported by IT Home

Reportedly: Alibaba launches HappyShrimp AI music creation app with subscription starting at RMB 30 per month

Alibaba's ATH team launched HappyShrimp, an AI music creation app on iOS and Android, IT Home reported. The app generates lyrics, composition, arrangement, and vocals from natural…

Why it matters Alibaba's entry into consumer AI music creation puts a direct competitor in front of existing AI music tools, potentially pressuring their pricing and feature sets.

NEW · Meituan · Official source

Meituan LongCat open-sources LongCat-Video-Avatar-1.6 for audio-driven human video generation

Meituan LongCat released LongCat-Video-Avatar-1.6, an open-source audio-driven human video generation model, on Hugging Face under an MIT license. It improves lip sync, supports v…

Why it matters Developers outside China can now use an open-weights model that outperforms HeyGen and Kling Avatar 2.0 in human preference, reducing reliance on commercial avatar APIs.

NEW · DeepSeek · 3 sources

DeepSeek releases Family Bucket application API enabling 1:1 replication of its app

DeepSeek opened its Family Bucket application API on November 13, 2026, letting developers enable deep-thinking search, file input, and other app features via parameters in existi…

Why it matters Developers outside China can now build applications with DeepSeek's search, file parsing, and voice capabilities without managing tool orchestration, reducing integration cost and time to market for…

Nov 12, 2026 8 stories

NEW · Ministry of Industry and Information Technology · Reported by Zhidongxi

Reportedly: Beijing releases 2026-2028 embodied AI robot action plan targeting 200,000 units annual capacity and 50,000 PFlops dedicated compute

Beijing Municipal Bureau of Economy and Information Technology released the Embodied AI Robot Industry High-Quality Development Action Plan (2026-2028) on November 11, 2026. The p…

Why it matters Humanoid robot manufacturers in Beijing will face new policy-driven targets for production capacity and component localization, potentially accelerating investment and consolidation in the…

NEW · DeepSeek · Official source

DeepSeek open-sources chat-demo, a static browser chat client for its API

DeepSeek released chat-demo, an open-source static chat client that connects directly to the DeepSeek API from the browser. It is MIT-licensed and available on GitHub. Users can r…

Why it matters Developers outside China can now self-host a ready-made chat interface for DeepSeek's API, reducing integration effort and potentially lowering the barrier to adopting DeepSeek models.

NEW · ByteDance · Reported by IT Home

Reportedly: ByteDance Volcano Engine releases Seedance 2.5 flash video model: generation speed doubled, price lower

ByteDance Volcano Engine released Seedance 2.5 flash, a video generation model with doubled speed and lower price. It costs about 0.99 yuan/second for 720P text-to-video and image…

Why it matters ByteDance's Seedance 2.5 flash lowers the cost of high-quality video generation for content creators and businesses, making batch production more affordable and potentially pressuring competitors to…

NEW · DeepSeek · Reported by IT Home and ifanr

Reportedly: Doubao Work adds DeepSeek V4.1 Flash, Kimi K3.1 Pro, GLM 5.3 as third-party models with quota consumption multipliers and limited-time discounts

Doubao Work added DeepSeek V4.1 Flash, Kimi K3.1 Pro, and GLM 5.3 as third-party model options, IT Home reported. The update introduces quota consumption multipliers shown in the …

Why it matters ByteDance's Doubao Work now lets users choose third-party models like DeepSeek and Kimi, which changes the competitive dynamic for Chinese AI model providers by making their models directly…

NEW · ByteDance · Reported by Leiphone and Zhidongxi

Reportedly: Doubao Work adds Kimi 3.1 Pro, GLM 5.3, DeepSeek V4.1 Flash and launches limited-time quota discounts

Doubao Work integrated third-party models Kimi 3.1 Pro, GLM 5.3, and DeepSeek V4.1 Flash on November 12, Leiphone reported. It also launched limited-time discounts, including redu…

Why it matters ByteDance's Doubao Work now offers access to competing models from Moonshot AI, Zhipu AI, and DeepSeek, which could reduce reliance on ByteDance's own models for enterprise users and intensify…

NEW · Tencent · Reported by Pandaily

Reportedly: Tencent Launches Yuanbao Search, a Standalone AI Search Product at so.qq.com

Tencent launched Yuanbao Search, a standalone AI search product at so.qq.com and in QQ Browser, providing AI answers over the web, WeChat Official Accounts, Channels, and academic…

Why it matters Tencent's entry into standalone AI search directly challenges existing AI search products by leveraging its exclusive access to WeChat's vast content ecosystem, potentially shifting user search…

NEW · AgiBot · Reported by Pandaily

Reportedly: AgiBot Open-Sources LIFE, a Lifelong Embodied AI Framework Built Around RoboBeats Models

AgiBot open-sourced LIFE, a lifelong embodied AI framework using RoboBeats models for continuous sensing, reanalyzable memory, and reusable tools. Its tests show 60.5% on SSS-Benc…

Why it matters Robotics developers outside China gain access to an open-source framework for lifelong embodied AI, potentially reducing the cost of building robots that learn continuously.

NEW · Huawei · Reported by GeekPark and IT Home

Reportedly: ByteDance plans to launch a Meta Muse-like AI assistant within weeks

ByteDance is planning to launch an AI assistant similar to Meta Muse within weeks, Bloomberg reported, citing people familiar with the matter. The assistant aims to consolidate By…

Why it matters ByteDance's entry into the AI assistant space could intensify competition for consumer AI assistants, potentially affecting user acquisition costs and feature parity for existing players like Meta…

Nov 11, 2026 1 story

NEW · ByteDance · Reported by IT Home

Reportedly: ByteDance plans to launch an AI assistant similar to Meta's Muse within weeks

ByteDance plans to launch an AI assistant similar to Meta's Muse within weeks, IT Home reported, citing people familiar with the matter. The assistant would consolidate ByteDance'…

Why it matters ByteDance's reported move into voice-interactive AI assistants could intensify competition with Meta's Muse in consumer AI, potentially affecting user acquisition and hardware integration strategies…

View All events

New Chinese AI models

All models

Explore China AI

Your reading list

Save stories. Free, across devices.