AI News on July 2, 2026

Industry News

Meituan Launches LongCat-2.0: A Trillion-Parameter Model Trained on 50,000-Card Domestic Computing Clusters

Meituan's technology team has officially announced the release of LongCat-2.0, a groundbreaking trillion-parameter large language model. This release marks a significant milestone as the industry's first model of this scale—boasting 1.6 trillion total parameters—to complete its entire training and inference lifecycle on a domestic computing cluster featuring 50,000 cards. LongCat-2.0 was pre-trained from scratch and features native support for an ultra-long context window of 1 million tokens. Specifically engineered for "Agentic Coding" tasks, the model is designed to enhance efficiency and stability in code understanding, generation, and execution. With an average activation of approximately 48B parameters and a dynamic range of 33B to 56B, LongCat-2.0 represents a major leap in domestic AI infrastructure and specialized software engineering capabilities.

美团技术团队•July 2, 2026

Read AnalysisSource

Industry News

Meituan Technical Team Showcases Research Excellence with Selected Papers at ICML 2026

The Meituan Technical Team has announced the selection of its academic papers for the International Conference on Machine Learning (ICML) 2026. As one of the most influential global platforms in the machine learning field, ICML focuses on addressing future challenges and core issues within the industry. The conference prioritizes research that demonstrates significant theoretical value and practical impact, aiming to drive the development of the field and lead future research directions. Meituan's participation underscores its commitment to high-level academic contribution and the exploration of cutting-edge machine learning solutions. This selection highlights the team's role in contributing to the global academic discourse and its focus on research that balances theoretical innovation with real-world application.

美团技术团队•July 2, 2026

Read AnalysisSource

Research Breakthrough

LongCat Open Sources VitaBench 2.0: A Pioneering Benchmark for Long-Term Dynamic User Modeling in AI Agents

The Meituan technical team has officially open-sourced VitaBench 2.0, a groundbreaking benchmark developed under the LongCat project. This new framework is the first of its kind to focus on long-term dynamic user modeling within real-life scenarios. VitaBench 2.0 is designed to systematically evaluate the capabilities of Large Language Models (LLMs) in maintaining personalization and demonstrating proactivity throughout extended, evolving interactions. By shifting the focus from static, short-term tasks to complex, real-world user relationships, VitaBench 2.0 sets a new standard for the industry. It provides a rigorous methodology for assessing how AI agents adapt to user needs over time, ensuring that the next generation of AI is not only reactive but also deeply personalized and capable of taking initiative in dynamic environments.

美团技术团队•July 2, 2026

Read AnalysisSource

Open Source

Meituan Open Sources AIGC Poster Generation System Featuring a Complete Generation-Editing-Evaluation Technical Closed Loop

Meituan's Intelligent Creation Team has officially unveiled a comprehensive technical system for AIGC poster generation, marking a significant milestone in automated visual content creation. The system is built upon a sophisticated "Generation-Editing-Evaluation" closed-loop framework, designed to streamline the creative workflow from initial concept to final quality assurance. Currently implemented across Meituan Waimai (food delivery) and various brand IP scenarios, the technology demonstrates high practical utility in high-volume commercial environments. In a move to support the broader developer community, Meituan has fully open-sourced this technical architecture, providing a robust foundation for further innovation in the field of intelligent design and automated marketing materials.

美团技术团队•July 2, 2026

Read AnalysisSource

Research Breakthrough

Meituan LongCat Team Launches WBench: The First Multi-Round Benchmark for Interactive Video World Models

The Meituan LongCat team has officially introduced and open-sourced WBench, a pioneering systematic multi-round evaluation benchmark specifically designed for interactive video world models. Positioned as a diagnostic tool analogous to a "CT scanner," WBench is engineered to pinpoint the technical limitations encountered by AI models as they transition from passive video observation to active, multi-turn interaction. By providing a structured framework for assessment, WBench aims to clarify the boundaries of current world models, offering the research community a precise method to identify where models fail in maintaining consistency and responsiveness during interactive tasks. This development represents a critical advancement in the standardization of world model evaluation, focusing on the complexities of dynamic, user-driven environments.

美团技术团队•July 2, 2026

Read AnalysisSource

Industry News

Meituan Showcases AI Innovation at ACL 2026: Advancing LLM Evaluation, Reasoning, and Generative Recommendations

The Meituan technical team has announced the acceptance of six research papers at ACL 2026, a premier international conference in computational linguistics and natural language processing (NLP). These papers represent Meituan's latest breakthroughs in building a new paradigm for generative AI. The research spans five critical domains: large model evaluation, complex process reasoning, competition-level mathematical thinking optimization, reinforcement learning (RL) optimization, and generative recommendation systems. By focusing on these high-impact areas, Meituan aims to bridge the gap between theoretical AI capabilities and practical, real-world applications. This selection highlights Meituan's strategic investment in enhancing the intelligence, reasoning depth, and efficiency of AI models within its vast service ecosystem.

美团技术团队•July 2, 2026

Read AnalysisSource

Open Source

LongCat-Video-Avatar 1.5: Meituan Open-Sources Commercial-Grade Digital Human Video Model for High Fidelity and Stability

Meituan's technical team has officially open-sourced LongCat-Video-Avatar 1.5, a significant upgrade in digital human video generation designed to bridge the gap between experimental research and commercial-grade application. This latest iteration introduces comprehensive improvements in lip-sync accuracy, physical plausibility, and stability during long-form video generation. Furthermore, the model now supports complex multi-person interactions and features optimized inference efficiency. By focusing on reliability in complex commercial environments, LongCat-Video-Avatar 1.5 aims to transition digital human technology from controlled laboratory settings to diverse, real-world professional stages, offering high-quality, natural video output for a wide range of users.

美团技术团队•July 2, 2026

Read AnalysisSource

Research Breakthrough

LARYBench: Redefining Embodied Action Representation Through Large-Scale Human Video Learning

The Meituan Technical Team has introduced LARYBench (Latent Action Representation Yielding Benchmark), a systematic evaluation framework designed to guide the development of general latent action representations from massive visual datasets. This benchmark serves as a critical milestone, often compared to an 'ImageNet' for embodied actions. The research findings reveal a significant shift in AI development: general-purpose vision models demonstrate superior performance in action generalization and control precision when compared to specialized embodied AI expert models. Most notably, the study confirms that embodied action representations can naturally emerge from large-scale human video data, suggesting that the vast library of human motion can be a primary source for training sophisticated robotic control systems without the need for exclusive robotic telemetry.

美团技术团队•July 2, 2026

Read AnalysisSource

Industry News

Meituan LongCat Team Launches General 365: A Challenging New Benchmark for AI Reasoning

The Meituan LongCat team has officially released General 365, a new benchmark designed to evaluate the reasoning capabilities of large language models. In a comprehensive assessment of 26 mainstream models, the results highlight a significant gap in current AI reasoning performance. Gemini 3 Pro, currently regarded as one of the most capable models, achieved a top score of only 62.8%. Most other models failed to reach the 60% accuracy threshold, which the team identifies as the 'passing mark.' This release establishes a more rigorous standard for the industry, suggesting that complex reasoning remains a major hurdle for even the most advanced artificial intelligence systems.

美团技术团队•July 2, 2026

Read AnalysisSource

Industry News

Managing 310,000 Lines of Code Refactoring: Meituan’s Strategy for AI Coding via Agent Evaluation Thinking

Meituan's technical team has shared a comprehensive case study on refactoring 310,000 lines of code using AI. The core insight is that when AI generates over 90% of a system's code, the primary challenge shifts from development speed to the implementation of effective constraints. Without a unified framework, AI-driven development can lead to significant technical debt and system chaos. Meituan addressed this by adopting an "Agent Evaluation" mindset, focusing on technical debt sorting, rule establishment, a standardized refactoring SOP, and a Pre-PR mechanism. This shift has allowed the team to move away from high-cost, one-off refactoring projects toward a model of continuous, daily iterative improvement, ensuring that code quality remains high even as AI takes over the majority of the writing process.

美团技术团队•July 2, 2026

Read AnalysisSource

Open Source

AI-Berkshire: A Value Investment Research Framework Powered by Claude Code and Multi-Agent Analysis

AI-Berkshire is an innovative open-source project hosted on GitHub that bridges the gap between traditional value investing and modern artificial intelligence. Built specifically for Claude Code and Codex, the framework integrates the investment philosophies of legendary figures like Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu. By utilizing multi-agent parallel research and adversarial analysis, the project aims to automate and enhance the depth of financial research. This framework represents a significant shift in how investors can leverage large language models (LLMs) to apply rigorous, time-tested investment principles in the AI era, providing a structured approach to identifying value in complex markets through automated, high-level reasoning.

GitHub Trending•July 2, 2026

Read AnalysisSource

Open Source

Video-Use: Leveraging Coding Agents for Automated Video Editing via New Open-Source GitHub Project

Video-use, a project developed by the browser-use team and recently featured on GitHub Trending, introduces a specialized framework for editing videos through the application of coding agents. The project aims to shift the paradigm of video production from manual graphical interfaces to programmatic, agent-driven workflows. By utilizing intelligent agents capable of executing code-based instructions, video-use provides a method for automating complex video manipulation tasks. This development highlights a growing trend in the intersection of artificial intelligence and multimedia, where autonomous agents are increasingly used to streamline creative processes. The project's emergence on open-source platforms suggests a move toward developer-centric tools that prioritize scalability and automation in the video editing industry.

GitHub Trending•July 2, 2026

Read AnalysisSource

Industry News

Understanding Autoresearch and the Feedback Loop Behind Self-Improving AI Agents with Introspection Co-Founder Roland Gavrilescu

In a recent discussion, Roland Gavrilescu, the co-founder of Introspection, detailed the emerging paradigm of "autoresearch" and its role in the development of self-improving AI agents. The conversation highlights the technical framework of agent "recipes" and the implementation of self-improving loops that allow AI systems to refine their performance over time. A significant portion of the analysis focuses on the concept of the "software factory," where automation and AI-driven processes are becoming standard. Despite the high level of automation discussed, Gavrilescu emphasizes that humans remain a central and indispensable part of this software factory, providing the necessary oversight and direction for these self-improving systems. This insight provides a glimpse into the future of autonomous software development and the evolving relationship between human engineers and intelligent agents.

Latent Space•July 1, 2026

Read AnalysisSource

Industry News

Apple Reportedly Developing Redesigned Entry-Level MacBook Pro for 2027 and New iPad Pro Models

Apple is reportedly working on a significant update to its hardware lineup, with a "revamped" version of the entry-level MacBook Pro slated for a potential launch in the first half of 2027. According to reports from Bloomberg, this redesign marks a major shift for the base model of Apple's professional laptop series. Simultaneously, the company is in the testing phase for four new iPad Pro models. These tablets are expected to debut in the spring, with a primary focus on internal improvements rather than external design changes. The dual-track development highlights Apple's strategy of balancing aesthetic overhauls for its laptops with performance-driven updates for its high-end tablet category, ensuring both product lines remain competitive in the evolving tech landscape.

The Verge•July 1, 2026

Read AnalysisSource

Product Launch

ZCode Unveils GLM Coding Lite: A New Subscription Tier for Lightweight AI-Powered Development Workloads

ZCode has officially introduced "GLM Coding Lite," a specialized subscription tier designed specifically for developers managing lightweight workloads and small repository iterations. Priced at a competitive $16.2 per month—discounted from the standard $18—this plan includes a base usage allowance and offers rolling access to the latest flagship models and features. A significant highlight of the offering is its extensive compatibility, supporting over 20 coding tools alongside deep integration with the ZCode ecosystem. By targeting small-scale development and iterative coding tasks, ZCode aims to provide a cost-effective entry point for high-performance AI assistance, ensuring that developers working on smaller projects can still leverage the power of the GLM-5.2 harness and flagship model updates without the financial overhead of enterprise-level plans.

Hacker News•July 1, 2026

Read AnalysisSource

Product Launch

Qualcomm Linux 2.0 Launch: Empowering Developers with an Open and Unified IoT Ecosystem

Qualcomm has officially announced the release of Qualcomm Linux 2.0, a major update designed to transform the landscape of Internet of Things (IoT) development. This latest iteration focuses on two core pillars: openness and unification. By providing an open-source foundation and a unified development environment, Qualcomm aims to simplify the complexities associated with building and scaling IoT solutions. The release marks a strategic shift toward reducing fragmentation in the developer experience, allowing for more efficient creation of connected devices. As the industry moves toward more integrated hardware and software solutions, Qualcomm Linux 2.0 stands as a central platform for developers seeking a cohesive and transparent framework for their next-generation IoT projects.

Hacker News•July 1, 2026

Read AnalysisSource

Industry News

Elon Musk Denies Reports of SpaceX AI Phone Prototype Following Record-Breaking Initial Public Offering

Elon Musk has officially refuted a report from The Wall Street Journal claiming that SpaceX developed and showcased an AI-powered phone prototype. The report suggested that the aerospace company presented a "handset-like prototype" to potential investors shortly before its historic initial public offering (IPO) in June. According to the claims, the device featured a design that was notably slimmer than Apple's iPhone. However, Musk has dismissed these assertions as "utterly false," distancing SpaceX from rumors of an entry into the consumer smartphone market. The denial comes at a critical time for the company, following its successful transition to a public entity and amid intense speculation regarding its future technological ventures beyond aerospace and satellite communications.

The Verge•July 1, 2026

Read AnalysisSource