EastVoice · AI · 2026-09-23 · Issue #76

Editor's Note: Alibaba's Yunqi conference put the whole stack on one stage in a single day: Qwen4 in training, 5 to 10 trillion parameters planned for later versions, an AgentCore platform, enterprise digital employees, and a ¥899 desktop robot. OpenAI and Anthropic cut prices within hours of each other with GPT-6 Sol, GPT-6 Luna and Claude Opus 5.5. DeepSeek is heading to the UN Security Council to brief members on AI risk.

Commentary: Chinese labs spent the day shipping hardware and platform plumbing while the price war went global. The louder signal is what nobody claimed. No new capability frontier was announced, only cheaper tokens and more surfaces to put an agent on.

---

I. AI Foundation Models

1. OpenAI Ships GPT-6 Sol and Luna, Cutting API Prices by Half

• Summary: OpenAI released GPT-6 Sol and GPT-6 Luna on September 23, trained along the same path as GPT-6 Astra, with API prices reported down roughly 50% and fewer errors on factual and coding tasks.
• Source: ITHome · 2026-09-23
• Editor's Take: Astra is barely a month old and OpenAI is already filling in the cheap seats. Price bands, not benchmark wins, are what close enterprise deals.

2. Anthropic's Opus 5.5 Reaches Fable 5.1 Territory at 40% Lower Cost

• Summary: Anthropic launched Claude Opus 5.5 on September 23, the first model in its 5.5 family. The company claims performance comparable to Claude Fable 5.1 on most work at roughly 40% below Opus 5 running costs.
• Source: ITHome · 2026-09-23
• Editor's Take: A new family's first release anchors what buyers expect to pay for everything after it. Anthropic set that anchor deliberately low.

3. DeepSeek to Brief the UN Security Council on AI Risk

• Summary: Reuters reported on September 22 that DeepSeek / 深度求索 will appear before the UN Security Council this week to brief members on AI safety risks, as world leaders gather in New York for the General Assembly.
• Source: ITHome · 2026-09-22
• Editor's Take: A Chinese lab briefing the Security Council reframes AI safety as a multilateral file rather than a US export-control file. That shift matters more than any model shipped this week.

4. DeepSeek Chases 8 Trillion Parameters

• Summary: According to The Information, DeepSeek is training a model of roughly 2 trillion parameters, and founder Liang Wenfeng / 梁文锋 told investors recently that the next target is 8 trillion.
• Source: Huxiu · 2026-09-22
• Editor's Take: DeepSeek built its reputation on efficiency and is now entering the parameter race it once dismissed. Scale is a strategy only if you can serve it at a price.

5. Alibaba Confirms Qwen4 Is Training, 5 to 10T Parameters Next

• Summary: At Yunqi on September 22, Alibaba said Qwen4 is being trained on a new architecture and that Qwen4.5 and Qwen5 will scale into the 5 to 10 trillion parameter range. Recursive self-improvement has already reached training, inference and chip-model co-design.
• Source: QbitAI · 2026-09-22
• Editor's Take: Naming the next three versions and their parameter band is a supply-chain signal, aimed at whoever is committing GPU budget for 2027.

6. A $15.6B US Legal AI Unicorn Turns to Chinese Open Models

• Summary: Harvey, the legal AI company valued at $15.6 billion, is rethinking its reliance on OpenAI as inference and compute costs climb, and has been evaluating Chinese open-weight models, per a September 22 report.
• Source: Huxiu · 2026-09-22
• Editor's Take: Cost pressure at the application layer is doing the work no procurement memo could. Chinese open weights just got their first serious US enterprise reference account.

7. Inspur Fits a 2.8T-Parameter Kimi K3 on One Rack

• Summary: Inspur unveiled the Yuanbrain SD200 Ultra supernode at AICC 2026, claiming a single unit can serve Kimi K3 at 2.8 trillion parameters, with token latency down to 5.85 milliseconds.
• Source: ITHome · 2026-09-22
• Editor's Take: The 5.85 ms figure matters more than the parameter count. Frontier Chinese models no longer need a cluster the customer has to assemble first.

8. Loongson Ships the Software Stack for Its Own GPUs

• Summary: Loongson / 龙芯中科 released the first software version of its accelerated computing platform for in-house general-purpose GPUs, covering drivers, a programming environment, OpenCL 3.0, CUDA compatibility and AI inference.
• Source: ITHome · 2026-09-22
• Editor's Take: Driver and compiler support is the part nobody budgets for and everybody gets blocked on. This is a supply-chain hedge, not a benchmark entry.

---

II. AI Software & Applications

1. Alibaba Cloud Rebuilds Itself Around Agents

• Summary: At Yunqi on September 22, Alibaba Cloud CTO Li Feifei / 李飞飞 laid out an "Agentic Cloud" strategy built on models, harnesses and context, with new products including AgentCore, Agent Sandbox and the next-generation CPFS storage system.
• Source: Leiphone · 2026-09-22
• Editor's Take: Treating harness and context as first-class cloud primitives is the clearest statement yet that Chinese cloud vendors expect agents, not people, to be their main API consumer.

2. Qwen Goes After a Personal Agent for Everyone

• Summary: Qwen said at Yunqi that it is accelerating work on a Personal Agent built on the agentic capabilities of the Qwen 3.8 series, with the stated goal that every user gets their own AI assistant.
• Source: Leiphone · 2026-09-22
• Editor's Take: Consumer agents live or die on the phone's default assistant slot. Alibaba is arguing for that slot through hardware rather than through the app store.

3. QwenWork Ships Enterprise Context, Digital Employees and a New Box

• Summary: Alibaba's QwenWork released enterprise agent products on September 22, including enterprise context, digital employees and the QwenNote A2 agent hardware, pitched at agents that understand the business, collaborate with staff and can be trusted.
• Source: ITHome · 2026-09-22
• Editor's Take: Enterprise context is the quiet layer of every agent deployment. Whoever owns it owns the switching cost.

4. Honor's Magic9 Will Debut Alibaba's Qwen Intelligence on September 28

• Summary: Alibaba launched Qwen Intelligence, a full-stack AI phone solution, with Honor's Magic9 series, due September 28, as the first device to carry it. Alibaba calls it the industry's first system-level Agent Harness architecture.
• Source: ITHome · 2026-09-22
• Editor's Take: A system-level harness puts the agent below the app layer, where a user cannot uninstall it. Google and Apple are fighting the same battle.

5. Tencent's Hunyuan Adds a Professional Image Model

• Summary: Tencent released Hy Image3.5 preview on September 22 with text-to-image, image-to-image and multi-turn creative dialogue, shipping into Yuanbao and ima. Tencent Video's WorkRally platform opened it to creators free for two weeks.
• Source: ITHome · 2026-09-22
• Editor's Take: Two free weeks for film and series production teams is how reference customers get made. Generated storyboards are the cheapest way into a studio's pipeline.

6. China Unicom Turns Its Hotline Into an Agent

• Summary: China Unicom released AICC3.0, moving its intelligent hotline from AI-assisted answers to autonomous task execution, so a call can finish a job instead of only explaining one.
• Source: ITHome · 2026-09-22
• Editor's Take: Government and enterprise hotlines are agent deployments with a real success metric. Completion rate, not satisfaction score, is the number to watch.

7. China Nears 5 Million Active Agents, Compute Market to Triple by 2030

• Summary: The 2026 China AI Computing Power Development Assessment Report put active agents in China at nearly 5 million, projecting compound annual growth above 150% from 2026 to 2030 and a 300% expansion in the AI compute market.
• Source: ITHome · 2026-09-22
• Editor's Take: Agent counts stay a vanity metric until someone reports how many survive past week one. The compute forecast is the number utilities and chip vendors actually buy against.

8. Alibaba's Qwen Glasses Add Iris Payment

• Summary: Alibaba showed the Qwen AI Glasses N1 and N1 Pro at Yunqi alongside clip-on AI earbuds. The glasses support eye tracking and iris payment, and go on sale October 13.
• Source: ITHome · 2026-09-22
• Editor's Take: Iris payment makes the glasses a payment terminal, which changes who has to approve them before they ship. Payments compliance is now a wearable supply-chain problem.

9. Forbes: AI Built 7 of America's 10 Youngest Billionaires

• Summary: Forbes' annual list of the 400 richest Americans put the entry bar at a record $4.4 billion. The number of billionaires under 40 rose from 4 to 10, and 7 of them built their fortunes on AI, four of them at Anthropic.
• Source: ITHome · 2026-09-22
• Editor's Take: Four of those ten sit inside one company. Equity concentration at this scale shows where the returns of this cycle are landing.

---

III. Humanoid Robots

1. Tesla Audits Chinese Suppliers for Optimus Production

• Summary: Tesla managers and engineers have begun auditing Chinese suppliers including Sanhua Intelligent Controls, as the company works out how to mass-produce Optimus while Washington tightens restrictions on Chinese robotics.
• Source: Huxiu · 2026-09-22
• Editor's Take: Washington writes the export rules and Tesla shops the supply chain anyway. The humanoid bill of materials still runs through China.

2. Tesla Orders 5,000 Optimus Units, Team Lands in Ningbo

• Summary: Tesla placed orders for roughly 5,000 Optimus robots with suppliers and sent a team to Ningbo on September 16 to audit robot suppliers, according to supply-chain reports.
• Source: TMTPost · 2026-09-22
• Editor's Take: The distance from one robot to ten thousand is manufacturing engineering, and the car industry already has it. That is why Tesla keeps beating robotics startups on cost.

3. Chinese Humanoid Sector Faces a Turnaround or a Shakeout

• Summary: A public dispute between Mekamand founder Shao Tianlan / 邵天兰 and Galaxy General / 银河通用 over alleged fabricated revenue, related-party transactions and a rushed listing has exposed how thin the commercial floor is under parts of China's embodied AI sector.
• Source: Huxiu · 2026-09-22
• Editor's Take: When founders start accusing each other of booking fake revenue in public, the funding environment has already turned. The audit trail will matter more than the demo reel from here.

4. Qiyuan's ¥20,000 Robot and a ¥60 Billion Market Cap

• Summary: Qiyuan Robotics / 启元机器人, launched in late 2025, held a product event in Shanghai on September 20 with celebrity guests, and now has to justify its parent Shangwei New Materials' roughly ¥60 billion valuation with ¥20,000 consumer robots.
• Source: Huxiu · 2026-09-22
• Editor's Take: Pricing a robot like a home appliance sets a volume expectation the company cannot meet yet. The valuation is now a delivery schedule.

5. Japan Plans an AI Robot Contest Aboard the ISS

• Summary: Nikkei reported that Japan plans to hold a robot competition aboard the International Space Station as early as 2027, with teams using AI to control robots and the stated goal of training the next generation of space engineers.
• Source: ITHome · 2026-09-22
• Editor's Take: An orbital contest with a two-year runway is a talent pipeline dressed as a race. Robotics recruiting keeps starting earlier.

---

IV. Autonomous Driving

1. Mercedes-Benz Taps Wayve for Production Autonomy

• Summary: Mercedes-Benz agreed on September 22 to develop autonomous driving systems with UK startup Wayve, with at least one model carrying the system due within two years.
• Source: ITHome · 2026-09-22
• Editor's Take: Wayve's end-to-end stack just got a luxury OEM as its first real platform. For a company that bet against HD maps, German homologation is the proof point that counts.

2. Tesla's FSD 14.3.9 Flags Clean Cameras as Dirty

• Summary: Owners running FSD (Supervised) 14.3.9 report frequent alerts claiming the front camera is dirty and requiring a service visit, even when the windshield and camera housing are clean.
• Source: ITHome · 2026-09-22
• Editor's Take: A cleaning warning that fires on clean glass is a perception threshold problem, not a hardware one. False alarms at this rate send owners to service centers for nothing.

3. Czech Republic Clears Tesla FSD for Public Roads

• Summary: The Czech Republic formally approved Tesla's FSD (Supervised) system for use on public roads, adding another country to the European rollout list.
• Source: ITHome · 2026-09-22
• Editor's Take: Country-by-country approvals are the only route into the European market. Each one makes the next filing cheaper.

4. Robot Buses, Not Robotaxis, Decide the Transit Market

• Summary: A TMTPost analysis argues that while Robotaxi takes the headlines, fixed-route autonomous buses are the vehicles most likely to enter urban public transit at scale.
• Source: TMTPost · 2026-09-22
• Editor's Take: Fixed routes and depots remove the hardest part of autonomy, which is the long tail. Fleet economics get proven in the unglamorous segment first.

---

V. World Models / Physical AI

1. XPeng's Second-Gen VLA Moves From 3D Space to 4D Time

• Summary: XPeng began rolling out XOS 6.3.0 for its second-generation VLA system on September 22, describing a move from 3D spatial recognition to 4D spatiotemporal understanding through stronger temporal memory, trajectory prediction and decision response.
• Source: ITHome · 2026-09-22
• Editor's Take: Adding time to the representation is the same problem world-model teams are working on, shipped to a car first. Driving stays the fastest route for spatiotemporal models into production.

2. Alibaba Says Qwen4 and Its Next Video Model Are Both Training

• Summary: At Yunqi, Alibaba confirmed Qwen4 is in training and that its next-generation video model is also underway, alongside progress on a native omni-modal generation model the company expects within three years.
• Source: QbitAI · 2026-09-22
• Editor's Take: Video generation is where Chinese labs are placing their world-model bets, because the training data arrives free with the product. Whether that also buys physical understanding is still open.

---

物界前沿 | EastVoice