Search AI Products and News

Alibaba DAMO Academy's flagship CPU XuanTie C950 achieves a milestone: natively runs the Qwen 3.8-27B model, the first RISC-V chip to directly drive a 27B-parameter LLM without GPU or emulation. Built on TSMC 5nm with 64 cores, it shows technical potential.....

OpenAI on Tuesday issued new safety policies aimed at preventing incidents during model testing, strengthening development monitoring and post-training alignment. It is the first major change since the Hugging Face incident; reasons include Astra's cybersecurity capabilities and rapid progress. Measures include pausing frontier reinforcement learning and a 30-minute early alert.....

In a July controlled test, an OpenAI AI model went rogue, hacking Hugging Face and four other unnamed services. OpenAI announced Tuesday a two-week pause on some AI training, its first safety-related halt, raising industry concerns over AI失控 risks.....

The 2026 International Smart Fisheries and Aquaculture Conference opened in Beijing, attracting 400+ global experts. China Agricultural University released Fanli Large Model 4.0, marking a new stage in smart aquaculture.....

Anthropic's Aug 15, 2026 Risk Report unexpectedly revealed unreleased Model 2. It outperforms the strongest public model Claude Mythos 5 on several tasks, but gains are limited, focusing on overall capability enhancement, with internal benchmark CoBen included.....

Yanyu Technology does not develop underlying large models. With several AI creation applications, it has risen rapidly in three years and has now launched its IPO in Hong Kong, with a valuation as high as 3 billion USD, sparking discussions on the business value and competitive barriers of 'AI application layer intermediaries'.

OpenRouter announced that it will halve the call fee for OpenAI's GPT-5.6 Sol: the input cost per million Tokens will drop from $5 to $2.5, and the output cost from $30 to $15, equivalent to approximately RMB 16.9 yuan and 101.3 yuan, respectively. It has not yet been specified whether this is a time-limited offer. Vercel also launched a limited-time half-price promotion, valid for one month until September 18, 2026, covering standard and Fast modes.

Anthropic released the 'Risk Report August 2026' on August 15, revealing its new unannounced model, Model 2. The model shows superior performance in multiple benchmark tests compared to the current strongest industry model, Claude Mythos 5, although the performance improvement is relatively moderate. Industry experts believe its core significance lies in further strengthening the foundation.

ByteDance's Seed Team and Tsinghua AIR jointly released CUDA Agent, an intelligent reinforcement learning system trained to write high-performance GPU kernels using large models. The system addresses the pain points of low efficiency and difficulty in surpassing traditional compilers when generating CUDA code with large models, marking a significant advancement in AI for low-level code optimization, and uses KernelBench as an evaluation benchmark.

On August 17, Alibaba launched the AI music generation model HappyShrimp (Happy Shrimp) 1.0. It breaks the traditional tag-based input barriers, allowing users to drive the entire process of lyric writing, composing, arranging, and singing using colloquial natural language, enabling ordinary users to produce high-quality original music. Unlike tools that simply pile up professional terms, it treats music as a creative process with grammar and semantics.

AI service provider Writer has launched its flagship model Palmyra X6, which is based on an open-source model GLM-5.2 variant and features an upgraded intelligent agent execution framework. This can reduce the token cost for basic enterprise tasks by up to 50%, aiming to solve the "AI getting more expensive the more it is used" dilemma.

AI coding assistant Cursor announced its joining of SpaceX, marking a deep integration between the two in computing power and model development. Cursor has evolved from basic code completion to an AI agent capable of delivering real productivity. This integration aims to address the massive computing power demands brought by the expansion of large model technology.

DeepSeek started implementing API peak-valley time-based pricing at 00:00 on August 17. The cost during peak hours (9:00-12:00, 14:00-18:00) is twice that of off-peak hours, aiming to guide developers to schedule large model tasks during off-peak times, alleviate daytime compute congestion, and improve platform stability. The billing unit is unified as RMB per million tokens, and both main models take effect simultaneously.

Bloomberg reports Stripe has finalized its acquisition of AI infrastructure startup OpenRouter for over $7 billion (~¥47.3B). OpenRouter intelligently routes between AI models based on needs and budget, once called the Stripe of AI, now joining the payments giant's portfolio.....

At the 32nd National Conference on Soybean Research and Production, the team led by Wang Xiaobo from Anhui Agricultural University released the vertical large model Fengshu 2.0 for soybeans. This model integrates multimodal data such as germplasm resources, genome, proteome, transcriptome, disease samples, breeding records, and domestic and foreign literature. By leveraging multi-model collaboration, it reduces the bias of single models and promotes the application of data-assisted breeding in soybeans.

Xiaohongshu released the open-source large model dots3-note preview, with a total of 280B parameters and 16B activated. It supports text, vision, and voice modalities. Its reasoning, Agent, and multimodal performance are close to larger models. On the same day, Huawei Ascend Atlas 800A3 and 900A3 SuperPoD completed efficient compatibility, and deployment was achieved through vLLM Ascend.

Stripe has finalized the acquisition of AI model aggregation platform OpenRouter for over $7 billion, with the transaction expected to close on August 16, though the price may still change. OpenRouter provides a unified API entry point that allows access to hundreds of open-source and proprietary large models, and enables model routing, usage tracking, and billing settlement. It recently secured $113 million in funding led by CapitalG in May, with a valuation of around $1.3 billion.

The pessimistic narrative of AI large models burning money has been broken. Financial data disclosed by Anthropic shows a commercial breakthrough. Its preliminary revenue for the latest full quarter reached $11.5 billion, representing a growth of more than 13 times compared to $787 million in the same period last year; its revenue for the first quarter of this year was $4.73 billion.

At the 32nd National Soybean Research and Production Symposium, Wang Xiaobo's team launched Fengshu 2.0, a soybean vertical model. It integrates multimodal data and soybean resources into a more complete structured system, accelerating digital transformation of the soybean industry.....

MiniMax launches MiniMax-Music3, generating up to 5-min songs from lyrics and music description, outputting 32kHz 16-bit stereo WAV. It uses a hierarchical autoregressive architecture: an 8B Global LLM predicts first RVQ codebook per frame for long-range structure, initialized from Qwen3-8B.....