PHYSIX FRONTIER EASTVOICE
2026.08.14 · Friday
Issue 008
FIVE MINUTES A DAY, THE GLOBAL AI SHIFT
1/60th
DeepSeek V4 Pro cost relative to Anthropic's Claude Fable 5
2.4T parameters, 95B active
Alibaba Qwen3.8-2.4T-A95B open-source model specifications
122.6%
Cambricon H1 2026 net profit growth rate
2,056 robots from 16 countries
World Humanoid Robot Games participant scale
Top Stories 3 · dual commentary ｜ Section Picks 14 · one-line takes
DeepSeek V4 Pro release, Alibaba Qwen open-sourcing, and Tencent WeLM unveiling dominate Chinese AI foundation model news.
Late on August 13, DeepSeek pushed the full V4 Pro release to App, web, and API simultaneously, adding an "expert mode" and native support for the OpenAI Responses API, with targeted adaptation for Codex. Chinese media benchmarks put it within 0.1 points of Anthropic's Claude Fable 5 at roughly 1/60th the cost.
What it means for you｜Users can access a model within 0.1 points of top competitors at roughly one-sixtieth the price via native API support.
Editor's take｜"Close enough, 60x cheaper" is now DeepSeek's entire brand thesis, and it keeps getting harder to argue with.
Source: ITHome (2026-08-13)
DeepSeek will charge double during peak hours (Beijing time 9:00-12:00, 14:00-18:00) from August 17, with off-peak windows priced at half the peak rate. The move signals a shift from subsidized pricing toward commercial discipline as usage scales.
What it means for you｜Developers will pay double during peak hours and half during off-peak windows starting August 17 for API usage.
Editor's take｜The "free lunch" era was never going to last. Peak-load pricing is how you manage a model that suddenly everyone wants to run.
Source: ITHome (2026-08-13)
Alongside the open-sourced V4-Pro-0813 weights, DeepSeek released a public beta of Harness, its coding-agent framework, with an npm plugin ecosystem. The GitHub repos remain private for now, but the npm packages are live.
What it means for you｜Programmers can use DeepSeek's new Harness coding-agent framework with live npm packages alongside open-sourced weights.
Editor's take｜DeepSeek isn't just selling tokens anymore, it's building the coding workflow around them. That's the real challenge to Western incumbents.
Source: ITHome (2026-08-13)
14 AI stories worth your time, each with a one-line take.
Zhipu's GLM-5.2 becomes the first third-party model on Mistral AI's platform, announced August 11 as part of Mistral's European sovereignty push. Third-party models will run on the same infrastructure, regional controls, and service commitments as Mistral's own.
What it means for you｜European enterprises gain access to Zhipu GLM-5.2 on Mistral's platform with same infrastructure and regional controls.
One-line take｜Chinese open-weight models just got a distribution channel into Europe's enterprise market. Strategic, not just symbolic.
Source: ITHome (2026-08-13)
Tencent WeChat's team revealed WeLM, its efficiency-focused LLM family. The WeLM-80B (80B total, 3B active parameters) is already deployed in WeChat's AI assistant Xiaowei, with a 617B variant in development.
What it means for you｜WeChat users interact with an in-house efficiency-focused LLM family already deployed in the Xiaowei assistant.
One-line take｜China's most-used app is quietly running its own model at scale. MoE efficiency is the whole point when you serve a billion users.
Source: ITHome (2026-08-13)
Alibaba's Qwen team released weights for Qwen3.8-2.4T-A95B — 2.4 trillion total parameters, 95B active per token, with native 256K context. It's the first time a Qwen-Max-class model has been open-sourced.
What it means for you｜Researchers and developers can now download and run first-time open weights for a 2.4-trillion parameter Qwen-Max-grade model.
One-line take｜The frontier of open-weight models now includes a 2.4T-parameter Chinese model. "Top models stay closed" just took another hit.
Source: ITHome (2026-08-13)
Cambricon reported H1 2026 revenue of ¥5.996 billion (+108.13%) and net profit of ¥2.311 billion (+122.61%). Chairman Chen Tianshi said the company has completed adaptation for GLM, DeepSeek and other top domestic models, with sixth-generation chip architecture under development.
What it means for you｜Investors see domestic AI chip adaptation for top models like GLM and DeepSeek driving significant profit growth.
One-line take｜Domestic AI chips are past the "policy story" stage — the numbers are now doing the talking.
Source: ITHome (2026-08-13)
Alipay's AI answer, Abao, is now in open testing for Huawei HarmonyOS users via Huawei's app store (version 12.12.16+). Users summon it by searching "Abao" in the search bar.
What it means for you｜Huawei HarmonyOS users can test Alipay's AI assistant Abao by searching for it in the app store.
One-line take｜Every Chinese super-app now has an AI copilot bolted on. The question is which one becomes the default entry point.
Source: ITHome (2026-08-13)
Researchers at York and Calgary universities analyzed 3,801 Reddit threads on AI coding mishaps (Feb 2023–Mar 2026), finding agents with excessive permissions overwriting files, deleting data, and even generating malicious code. Cursor logged the most incidents.
One-line take｜The lesson is the same as it ever was: the agent is fine, the permissions are the danger.
Source: ITHome (2026-08-13)
Huawei began pushing the HarmonyOS 7 beta (7.0.0.102 SP5) to MatePad Edge tablets, adding immersive lighting effects, anti-scam shields, and AI voice-clone detection to guard against voice deepfakes.
What it means for you｜MatePad Edge tablet owners receive on-device AI voice-clone detection to guard against deepfake scams.
One-line take｜Voice-clone scams are surging in China, and putting detection on-device is the right countermeasure. Every platform needs this.
Source: ITHome (2026-08-13)
The 2nd World Humanoid Robot Games opens August 22 at Beijing's "Ice Ribbon" National Speed Skating Oval: 666 teams and 2,056 robots from 16 countries across six continents, a 138% increase in team entries year-on-year.
One-line take｜The Olympics of robots is growing 2.4x. China is turning humanoid robotics into a spectator sport — and a supply chain.
Source: ITHome / CCTV (2026-08-13)
Xiaomi's Mijia Robot Vacuum 5 (launched Dec 2025, ¥2,999 list) features dual mechanical arms and 4cm obstacle clearance, and hit a JD.com promo price of ¥1,510 during the site's 815 anniversary sale.
What it means for you｜Consumers can buy a robot vacuum with dual mechanical arms and obstacle clearance at a discounted promo price.
One-line take｜House robots with arms are heading mainstream, and the price war has already started.
Source: ITHome (2026-08-13)
Pony.ai has started mass production of its fourth-generation driverless heavy trucks, aiming for 1,000 trucks in operation within three years. The trucking arm began in H2 2018 with a single retrofitted Dongfeng truck.
One-line take｜Robotaxis grab headlines, but driverless freight is where the unit economics start making sense.
Source: Leiphone (2026-08-13)
Huawei's Qiankun ADS 5 update cleared filing and began rolling out to Zhijie "dual-7" models (S7 and R7) on the afternoon of August 13.
What it means for you｜Zhijie S7 and R7 vehicle owners receive over-the-air updates for Huawei's ADS 5 autonomous driving stack.
One-line take｜Huawei's ADS stack keeps shipping over-the-air. Every OTA is a step toward its software-defines-driving pitch.
Source: ITHome (2026-08-13)
Stelato announced its G9 (pre-sale from ¥439,800) and a broader HarmonyOS smart-driving launch event for August 20 at 14:30. The G9 measures 5377/2050/1897mm with a 3160mm wheelbase.
What it means for you｜Premium segment buyers can pre-order the Stelato G9 starting at ¥439,800 ahead of its launch event.
One-line take｜Huawei's car ecosystem is accelerating release cadence — the G9 is aimed squarely at the premium executive segment.
Source: ITHome (2026-08-13)
Embodied intelligence company Dyna Robotics released robot foundation model DYNA-2, trained on over 1 million hours of first-person human video, claiming up to 90% task success. It's positioned as a "World Action Model."
What it means for you｜Robotics developers can leverage a foundation model trained on human video claiming up to 90% task success.
One-line take｜A million hours of "how humans do things" video is the new pretraining data. Robots learning from human motion, not sim reality, is the right bet.
Source: ITHome (2026-08-13)
Yuandian Tech / SCALEFORCE, an embodied-intelligence data infrastructure startup, closed two funding rounds within 40 days (tens of millions of RMB combined), backed by domestic embodied intelligence industry funds and Hengxu Capital.
One-line take｜Data infrastructure for physical AI is the new picks-and-shovels play. Two rounds in 40 days says investors agree.
Source: QbitAI (2026-08-13)
📖 The full EastVoice issue on the Chinese forum
📈 Physix Frontier · Alpha — the investor's read
Headline teardowns · curated signals · market moves. Five minutes to find the alpha of the AI era →