Anthropic claims that the Claude web and desktop versions have rebuilt their streaming renderers, updating only content that is still changing, resulting in approximately 4 times smoother streaming of long answers. These performance figures are from their own official testing.
AI programming tool Command Code has announced the addition of Alipay as a payment option, providing users with a new subscription payment method.
The OpenCode head stated that the $5 starter offer was canceled due to widespread abuse, and the team is still evaluating a new offer that is less prone to abuse; Go's current standard price is $10 per month.
GMI Cloud will offer MiniMax M3, M2.7, Speech 2.8, and Music 3.0 for free from August 24 to September 6, accessible via GMI Cloud API or OpenRouter; H3 is not included in the free offer.
NVIDIA Groq 3 LPX has entered full production and is positioned as a low-latency token generation extension for Vera Rubin NVL72. Nebius has become the first AI cloud service provider to adopt this platform.
University students who have completed Alibaba Cloud student verification can redeem a 1-month Qoder CN Personal Professional Edition Pro voucher. Verification, voucher redemption, order placement, and login must all be done using the same Alibaba Cloud account; each person is limited to one redemption.
According to reports, ByteDance has integrated the TRAE and Kouzi teams into the Doubao system, and related work scenario capabilities will be further integrated. ByteDance confirmed that this adjustment is aimed at coordinating resources and stated that existing user rights will not be affected.
Xiaomi released three self-developed chips: the Xuanjie O3, O100, and D100, and showcased an AI Cube engineering prototype featuring a three-chip collaborative architecture. This device supports local deployment of models with up to 120 bytes of parameters, but mass production and pricing information are not yet available.
Reuters, citing The Information, reported that Nvidia is in talks to participate in a new round of equity financing for Perplexity, with a proposed valuation exceeding $30 billion. No public deal has yet been finalized.
Alibaba's Wan3.0 video generation model has officially launched, capable of generating videos up to 30 seconds long in a single session, and supporting input of documents such as doc, xls, ppt, pdf, and md. Some platform APIs are offering a 30% discount from August 24th to September 23rd.
Thinking Machines Lab has launched the Tinker grant program for research on the security of open weighted models, with a maximum of $50,000 in Tinker credits available to a single project.
Mistral AI and Saudi Arabia's HUMAIN announced a strategic collaboration worth hundreds of millions of euros, covering AI infrastructure, model localization and deployment, with a focus on exploring cybersecurity, voice and Arabic models.
NVIDIA has released early test data for the Vera Rubin NVL72, claiming that it can achieve up to 30 times the throughput per megawatt and up to 35 times the token cost reduction compared to the GB300 NVL72 under certain proxy coding workloads.
SpaceX plans to deploy NVIDIA Vera CPUs and expand its Grok infrastructure with Vera Rubin, while also planning its first-generation Starmind AI satellite using an optimized version of the Vera Rubin NVL72 system.
Wan3.0 has been launched on RunningHub, with models generating up to 30 seconds of video per session. It supports input of text, images, audio, video, and documents such as doc/ppt/pdf, enabling one-click PPT to video conversion. The consistency between people and scenes has been significantly improved, eliminating the "AI face." RunningHub's ecosystem, including rhTV, RHSTORY, and ComfyUI, is fully compatible, covering commercial scenarios such as film, advertising, and e-commerce.
Alibaba announced a placement of 710 million new shares at HK$112.70 per share, with a transaction size of approximately HK$80 billion. The net proceeds will be used entirely for the construction of full-stack artificial intelligence capabilities and AI infrastructure.
China Telecom's Tianyi AI has launched TeleAgent, a desktop version of its Starry Sky Super Intelligent Agent, positioned as an AI office assistant. It supports Windows 10 and macOS 13 and above. The tool can automatically complete tasks such as file organization, Excel summarization, data analysis, PPT creation, and data collection. Users can simply give voice or text commands to get the software to do what they say.
Tibo, head of OpenAI Codex, said the team has identified the additional overhead caused by long session image processing, high quantile usage in Computer History, and session title generation, and plans to launch a fix and reset the usage of all paid subscriptions.
DeepSeek's current API pricing rules show that peak billing only applies to the two UTC periods from Monday to Friday, while off-peak pricing applies for the rest of the time. Therefore, the off-peak pricing applies to the entire day on Saturdays and Sundays.
According to Business Insider, Hugging Face is assessing interest from potential buyers with a bank, with the deal potentially valued at over $13 billion; discussions are still in the early stages and no confirmed buyer or agreement has been reached.
According to Bloomberg, some of Nvidia’s major customers have been told that the price of AI servers equipped with Vera Rubin and Grace Blackwell may increase by more than 15% in many cases early next year due to rising memory chip costs.
Claude Code team members stated that recent testing changed the mapping method between internal values and effort levels, but the effort levels selected by users remained unchanged, and internal evaluations did not find any impact on model performance.
Thibault Sottiaux, head of OpenAI Codex, said that some users have recently had a lower cache hit rate than previously stable, which may lead to faster usage. The team is investigating.
The OpenAI product manager stated that a banked reset will be issued to all paid ChatGPT Work and Codex users.