AI model

On August 11th, the Chinese AI model DeepSeek experienced a widespread service outage, impacting its API, website, and mobile app. The incident sparked widespread complaints on Weibo under the hashtag #DeepSeekDown. DeepSeek reported the issue was identified and resolved between 21:55 and 23:39. DeepSeek gained popularity in early 2025, surpassing ChatGPT in US app downloads and reaching 30 million daily active users faster than any other app.

OpenAI has launched GPT-5, touted as a significant upgrade with expert-level performance in programming, mathematics, writing, healthcare, and visual perception. GPT-5 features an intelligent model, a deep reasoning engine, and a real-time router. Free and paid users will get access, with subscribers enjoying higher usage. GPT-5 demonstrates improvements in response time, accuracy, instruction following, and reduced hallucination. It excels in programming, creative writing, and healthcare. GPT-5 Pro surpasses o3-Pro for complex tasks, achieving state-of-the-art results in GPQA.

The Shenzhen Institutes of Advanced Technology has introduced HYPIR, a new AI model for image restoration and enhancement. This model rapidly restores images, generating 8K detail in approximately 1.7 seconds. HYPIR distinguishes itself through its ability to produce remarkably accurate image details and preserve text integrity, a common challenge in other restoration models. Potential applications include cultural heritage preservation, film industry revitalization, scientific research, and medical imaging, offering significant advancements over traditional editing software.

Elon Musk plans to revamp humanity’s knowledge base using Grok 3.5/4, addressing the issue of misinformation in current AI models. The model will rewrite the human knowledge repository, correcting errors and filling gaps for subsequent retraining. Grok is designed to identify and eliminate inaccuracies, aiming to be the ultimate knowledge curator and counter the rise of AI-generated fabricated content.

Alibaba-backed Quark has launched an AI model to assist students with college applications. The tool features comprehensive exam search, application report generation, and intelligent college selection. It provides personalized advice, generates detailed reports with strategic suggestions, and recommends application plans. Quark has significantly increased its computing power to support free access to this service, including a specialized knowledge base and live expert broadcasts.

DeepSeek has rapidly upgraded its R1 model to DeepSeek-R1-0528, achieving better performance across benchmarks and closing the gap with leading AI models. The new model has been quickly integrated into various products, demonstrating its immediate value. This enhanced version also significantly reduces “hallucinations,” improving reliability in areas like summarization and comprehension by an estimated 45-50%.

DeepSeek has launched an updated R1 model, DeepSeek-R1-0528, improving complex reasoning, long-text handling, and code generation. This advancement positions the model competitively against industry leaders. Integrated with enterprise-focused features like enhanced security and robust infrastructure, the model offers batch processing, model evaluation, and customized LLM building. The upgrade seamlessly connects to application development, empowering businesses with advanced AI capabilities, including Retrieval-Augmented Generation (RAG).

Alibaba open-sourced its modular Wan2.1-VACE video generation model (1.3B/14B parameters), featuring multimodal input support (text, images, video clips) and Lego-like customizable modules. The lightweight 1.3B version runs on consumer GPUs, democratizing AI video creation. Released on GitHub/Hugging Face, it has gained 330K+ downloads and 11K+ stars, becoming a leading open-source video generation framework.