智东西China AInews

Qwen3.7-Max IS OUT🚀 In 35h autonomous core tests with 1000+ tool calls, it maintained coherent reasoning, achieved a 10× speedup. It can: Build frontend prototypes+ complex multi-file projects Automate workflows with MCP+ multi-agent collaboration Support mainstream frameworks

Ant Group rolled out the #Alipay AI Developer Incentive Program. It offers monthly Token subsidies up to 5,000 CNY, free fees for individual AI payments and a 40% discount for enterprises. And 9 business scenarios are open. Full details coming in June.

⚡️Alipay debuts the world-first Token Pay, Agents run stores & wallets Token Pay:+70% subscription success+Claw Token on-demand top-up AI Pay:95% agent framework compatibility AI Collect: Links merchants to global Agent orders AI Wallet: Authorize Agent payments, full control

🇨🇳He Tingbo of Huawei: Why talk about Tau (τ) Scaling Law today? #Huawei #Semiconductor #ChipTechnology

Try it now👉qwencloud.com

⚡️Alibaba Cloud unveils #QwenCloud, a new AI portal for the Agent era. Unlike traditional cloud AI modules, it packages core capabilities into Skills & CLI for agents to parse commands & call autonomously. Complete Qwen multimodal models available; third-party models to follow.

🇨🇳🤖 #AgriRobots work across #XinjiangCotton fields! · 4-wheel independent drive | 360° on-the-spot turning · Adjustable wheel track & ground clearance (fits complex & intercropped farmland) · 100% electric, zero carbon emissions · One-click remote control via mobile app

🚨JUST-IN: Code Arena new rankings Qwen3.7-Max scores 1541, outperforming GPT-5.5, Gemini-3.5-Flash, GLM-5.1 and Kimi-K2.6. It ranks 2nd globally among LLM providers, only behind the Claude family. #Qwen #LLM

➤HuggingFace：huggingface.openbmb.com/model/openbmb/… ➤GitHub：github.com/OpenBMB/MiniCPM ➤ModelScope：modelscope.cn/models/OpenBMB… ➤  GitCode：ai.gitcode.com/OpenBMB/MiniCP… ➤ Modelers：modelers.cn/models/OpenBMB…

🚨BREAKING: ModelBest×Tsinghua×OpenBMB open-source MiniCPM5-1B 1B params, outperforms all sub-2B models on AA-Index INT4 quantized (0.5GB), powers AI desktop pets on endpoints The base model is pre-trained on ForgeTrain, AI-coded production framework,10% faster than Megatron🚀

🚨BREAKING: The 2026 ISCAS kicked off in Shanghai on the 25th. He Tingbo, Huawei Director and head of Semiconductor Business Group, unveiled the “Tao (τ) Law” in her keynote, China’s first original guiding principle for the global #semiconductor industry.

➤ HuggingFace 🔗 huggingface.co/collections/op…  ➤  ModelScope 🔗 modelscope.cn/collections/Op…

🚨BREAKING: ModelBest × Tsinghua × OpenBMB open-source BitCPM-CANN, China’s first Ascend-trained 1.58-bit edge LLM Available in 0.5B/1B/3B/8B sizes 6x lower VRAM cut vs BF16, retaining 90%-97.2% original performance. 60B models will soon run on phones to lift mobile AI power🚀

⚡️#Tencent ima Copilot fully open, no queuing needed! As an Agent with built-in knowledge base and memory, it reads and calls KB content without leaving the current page. ima Knowledge Hub adds Skill publishing, evolving Knowledge Plaza from a content to a capability platform.

By 2031, high-end chips under this law will match 1.4nm-level transistor density.

The τ Law advocates time scaling instead of geometric scaling. Huawei has designed and mass-produced 381 chips in six years via this principle. It will launch new Kirin chips this autumn adopting full logic folding tech for major performance boosts.

@Samking207 ByteDance is indeed the best in this field right now. Hope it can keep this momentum.

@LunchangG7603 I think Bytedance will win the whole video generation AI, while Alibaba wins the LLM AI

🚨BREAKING: Seedance 2.1 will get a full core model revamp! · ~20% boost in visuals/stability/detail · Optimized facial, limb and temporal coherence · Stronger Chinese prompt parsing for long scripts · Standalone API available alongside v2.0 for seamless switching #ByteDance

@AliceInfoAi I think so too.

@LunchangG7603 Digital humans feel real

#Meituan open-sources #DigitalHuman model LongCat-Video-Avatar 1.5🚀 Upgrades lip sync, physics realism, video stability, multi-interaction & i nference. Audio encoder updated to Whisper-large GRPO adopted for human preference alignment DMD distillation speeds up inference 15x

Github：github.com/meituan-longca… HuggingFace：huggingface.co/meituan-longca… Tech Report：github.com/meituan-longca… Project Page：meigen-ai.github.io/LongCat-Video-… Modelscope：modelscope.cn/models/meituan…