Meituan Denies Rumors About LongCat Model Leader's Resignation, Continues to Promote the Implementation of Domestic Computing Power in Trillion-Level Models

On July 27, in response to rumors that Pei Peng, the head of the LongCat team's foundation model at Meituan, was about to leave, Meituan officially stated that the information was false.

Public information shows that Pei Peng graduated from Peking University. After joining Meituan in 2023, he successively served as the pre-training lead of the base model team and the head of the LongCat team's foundation model, comprehensively leading the research and development of Meituan's large language model, multimodal model, and Agent intelligent model, and leading the development and application of the trillion-parameter large model LongCat-2.0.

Meituan released LongCat-2.0 on June 30, which is the first trillion-parameter large model in the industry to complete training and inference using domestic computing power throughout the entire process. The model uses a MoE architecture, with a total of 1.6 trillion parameters, and about 48 billion parameters activated per Token. It originally supports a super long context of 1M and is deeply compatible with mainstream Harness such as Claude Code. Meituan quickly clarified the personnel rumors, demonstrating its strategic perseverance and the stability of its core team in the development of self-controlled large models.

Tencent Yuan and Multimodal Models Merge: The Multimodal and Large Language Model Departments Combine, YAO Shunyu Oversees the Full Modal Upper Limit

According to a report on July 24th, Tencent announced on July 23rd that it would merge the Yuan Multimodal Model Department with the Large Language Model Department to form the Basic Model Department, led by Chief AI Scientist YAO Shunyu. This move aims to enhance R&D and collaboration efficiency, and strive to achieve the intelligence upper limit of full modal models. There were hints of this integration earlier, as YAO Shunyu had already taken charge of the Large Language Model team in December last year. Now, combining both sides means Tencent is concentrating resources to promote the deep integration of multimodal and language models, accelerating the construction of a new generation of unified basic models, and moving towards higher peaks of full modal intelligence.

Epoch AI Tests Three AI Text Detectors: Nearly 30% of Content Missed After Mimicking Human Writing Styles

Epoch AI research shows that mainstream AI text detectors can nearly perfectly identify ordinary AI-generated content, but when large language models deliberately mimic the writing styles of specific authors, their accuracy drops significantly, with scientific writing being the hardest to detect. The experiment tested three tools: Pangram, GPTZero, and Originality.ai. It used 495 human-written texts covering blogs, novels, and scientific writing (all created before ChatGPT existed), and found that style imitation can effectively evade detection.

OpenAI Talent Mobility: Former Researcher Tian Yonglong Joins Tencent, Focused on Visual Language Model Development

Tian Yonglong, a former researcher at OpenAI, has joined Tencent's Large Language Model Department, focusing on the development of visual language models. This move is seen as a key recruitment for Tencent to strengthen its multi-modal large model strategy, highlighting the intense competition for cutting-edge talent.

Meituan's Large Model Ecosystem Adjustment: Fully Restricting Doubao and Promoting Its Self-Developed LongCat System

Meituan has internally restricted the use of Doubao, a large model under ByteDance, requiring business teams to self-check their usage. If retention is needed, they must submit a necessity explanation and migration plan. This move shows that Meituan is accelerating its shift toward self-developed AI solutions and adjusting its infrastructure choices.

Anthropic Releases the Powerful Large Model Claude Sonnet 5: Performance Approaches the Flagship, Price Drops Significantly

Anthropic launches its new mid-to-high-end model, Claude Sonnet 5, focusing on cost-effectiveness. Its performance has significantly approached the flagship Opus series. The model features the strongest agent capabilities to date, enabling it to independently plan complex tasks, self-check outputs, and flexibly utilize external tools such as browsers and terminals. It performs exceptionally well in reasoning, programming, and knowledge tasks.