Alibaba Drops Qwen3.8, Open Sourcing Next Week
On August 3rd, Alibaba officially unveiled its next-generation base model, Qwen3.8, boasting a massive 2.4 trillion total parameters. The model shows dramatic leaps in coding and professional office (Cowork) capabilities. The Qwen3.8 API is already live on the Qianwen AI platform, seamlessly integrated with Alibaba’s newly launched Agent product, “Qianwen Office.” The Qwen3.8-Max version is slated for open source next week, alongside the Qwen3.8-27B.
Zhiyuan Chief Scientist Leaving? Official Response: Not Leaving Yet, Just an Internal Role Shift
On August 3rd, market chatter suggested that Luo Jianlan, Chief Scientist at robot maker Zhiyuan, had recently parted ways. Zhiyuan’s official response to the news: he hasn’t left; it’s simply an internal role adjustment.
That same day, Zhiyuan’s official website published its full updated partner roster. The latest lineup reveals a team drawn from heavyweights like Huawei, Google, Tencent, DJI, and Cambricon, with expertise spanning AI R&D, product engineering, global commercialization, and government-enterprise relations.
DeepSeek-V4-Flash Official API Lands on the National Supercomputing Grid
August 3rd news: DeepSeek-V4-Flash’s official API has just entered public beta. The National Supercomputing Internet wasted no time, syncing up the model’s API call and download services. Users can head to the official site, click into the model services section from the homepage, and plug into the API. Or, for the full Monty, hit the AI community, grab the model, and run the entire fine-tuning, deployment, and adaptation workflow in a trusted Notebook environment.
OpenAI’s Next-Gen Model Astra Cracks Ten Math Problems on a $2,000 Compute Budget
On August 1st, OpenAI published ten breakthroughs in mathematics and theoretical computer science, all achieved by an internal version of its next major model, Astra. Running the numbers at Sol API rates, the total token cost to find these solutions? A cool $2,000. The ten wins include tighter upper bounds on high-dimensional sphere packing densities, improved bounds for binary and spherical codes, proof of non-simple Sophie groups, debunking the Connes rigidity conjecture, new lower bounds in arithmetic circuit complexity, quantum parallel repetition theorems, proof of polynomial-factor approximation difficulty for the nearest vector problem, the Ehrhart volume conjecture, super-exponential lower bounds for multicolor Ramsey numbers, and extremal number conjectures.
Insider Shoots Down Moonshot AI Hong Kong IPO Rumors: “Not True”
On August 3rd, reports surfaced claiming Moonshot AI was planning to file for a Hong Kong IPO as early as this month, aiming to raise around $3 billion. An insider close to the matter told the news: “The rumor is false. The Series G round is still in progress.”
Korea’s AI Chip Startup DeepX Sees Valuation Skyrocket to $2.2 Billion
On August 3rd, reports indicate that South Korean AI chip design startup DeepX has seen its valuation climb to approximately 3.14 trillion KRW (about $2.2 billion) in its latest funding round—nearly tripling from its previous valuation. DeepX has already inked the first tranche of its Series D, pulling in 42 billion KRW from existing shareholders BNW Investment and DS Asset Management. The company is now in talks to secure up to 300 billion KRW in follow-on funding before September, which could push its post-money valuation to as high as 3.5 trillion KRW.
