Alibaba Cloud Announces Price Cuts for Qwen3-VL-Rerank Model
On September 1st, Alibaba Cloud announced that its Bailian model service platform will lower the billing rates for the Qwen3-VL-Rerank model in the Beijing region, starting at 10:13:53 AM Beijing time on August 31, 2026. The price for text input will drop from 0.7 yuan per million tokens to 0.5 yuan, while multi-modal input will see a steeper reduction from 1.8 yuan per million tokens down to 0.5 yuan.
Anthropic Signs $35 Billion Cloud Computing Deal with Nvidia-Backed Lambda
Reports indicate that Anthropic has signed a massive $35 billion cloud computing agreement with Lambda, a cloud service provider backed by Nvidia. The data centers involved in this deal are currently being developed by Hut8, a bitcoin mining firm that also doubles as a data center developer, in Nueces County, Texas. Sources familiar with the matter say Nvidia inked its own agreement with Hut8 just a few weeks ago to lock down the computing power at that facility.
Qwen Announces Upgraded Learning Features
On September 1st, Alibaba’s Qwen announced an upgrade to its learning capabilities. The new features include unit-based essay tutoring that syncs with school textbooks, in-depth explanations of middle school Chinese and math materials, and expanded photo-solving capabilities for high school and college-level math, physics, and chemistry problems. All of these features are being offered to users for free.
Tencent Hunyuan Launches Hy4 Preview Lite Model
On September 1st, the Tencent Hunyuan team announced the release of a lightweight version of its Hy4 preview model, compressing the weights from 1.5TB down to roughly 214GB. According to the team, the compressed model retains long-text comprehension capabilities nearly identical to the original, with multi-turn long-context retrieval performing at essentially the same level. Math skills see only a slight dip, but everything stays well within acceptable bounds. Everyday coding assistance, tool invocation, long-document processing, and routine Q&A are all well-covered by this version.
Zhipu Responds to “Only Post-Training” Criticism: Next-Gen Base Model in Progress, GLM-6.0 Aiming for “Self-Evolution”
According to National Business Daily, during the 2026 semi-annual earnings call held on the evening of August 31st, an analyst raised the market’s skepticism about Zhipu “only being good at post-training,” asking why the company has been relatively conservative on parameter scale lately. Zhipu founder Tang Jie responded that the next-generation model will still expand the base scale, while keeping a tight rein on activated parameters to avoid slowing down inference speeds and driving up costs. On the computing power front, Zhipu disclosed that it has achieved large-scale inference on 100,000-plus domestically-produced chips, with per-token inference costs dropping 80% compared to the start of the year. GLM-5.3-Flash is Zhipu’s first model to fully rely on domestic chip clusters for serving under ultra-large-scale real-world traffic. The company claims that compared to the initial baseline on the same hardware, the model’s end-to-end service performance has improved threefold. Tang Jie summed up one direction for the future GLM-6.0 as “self-evolution.” He noted, “This is a huge challenge. The biggest question is whether the model can judge for itself when to stop and when to correct itself. That’s the real crux—not simply scaling the model up. In future models, this will be a key research focus.”
Apple Accuses OpenAI of “Destroying Evidence,” Claims Former Engineer “Took Confidential Info and Used It at OpenAI”
Reports say Apple is escalating its legal battle against OpenAI. In new court filings, Apple alleges that OpenAI is “actively destroying key evidence,” and that former iPhone engineer Chang Liu, who jumped ship to OpenAI, “not only downloaded Apple’s confidential circuit schematics but also used them in his work at OpenAI.” The court document, filed on August 31st local time, is meant to support Apple’s motion for expedited pre-trial discovery. Apple accuses OpenAI of stealing trade secrets, while OpenAI has asked the court to dismiss the case entirely.
Manus Resumes Independent Operations
On September 1st, Manus, the general AI agent company, announced that it has officially resumed independent operations. The company stated that its founding team will continue to lead, staying firmly committed to product innovation and advancing general AI agents for users worldwide. For some users, this transition involves backing up and restoring data, as well as dealing with temporary access interruptions. If you’ve already backed up your data but haven’t restored it yet, please use the recovery portal to do so. There’s no deadline for restoration. Users who weren’t affected by these measures don’t need to take any action.
China Securities Index Co., Ltd. to Launch 4 New Indices Including the CSI AI Application Software Select Index
China Securities Index Co., Ltd. will officially release four new indices on September 2, 2026, including the CSI Artificial Intelligence Application Software Select Index. The lineup consists of two AI application thematic indices and two AI application software thematic indices, designed to offer the market a richer set of investment options.
