AI News on July 13, 2026

Research Breakthrough

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 pioneering trillion-parameter model. This release marks a significant milestone as the industry's first model to complete its entire training and inference lifecycle on a domestic computing cluster consisting of 50,000 cards. LongCat-2.0 features a total of 1.6 trillion parameters with a dynamic activation range between 33B and 56B, averaging 48B. Built from scratch, the model natively supports an ultra-long context window of 1 million tokens. Its architecture is specifically designed to optimize Agentic Coding tasks, aiming to provide high efficiency and stability in code understanding, generation, and execution. This development highlights a major step forward for domestic hardware capabilities in supporting massive-scale artificial intelligence models.

美团技术团队•July 13, 2026

Read AnalysisSource

Industry News

Meituan Technical Team Showcases Machine Learning Innovations at ICML 2026

The Meituan Technical Team has announced its participation in the International Conference on Machine Learning (ICML) 2026, presenting a curated selection of academic papers. As one of the most influential global forums for machine learning, ICML 2026 focuses on addressing the core challenges and future trajectories of the industry. Meituan's contributions emphasize the balance between theoretical excellence and practical application, aiming to drive the field forward. This involvement highlights Meituan's role in the global research community, focusing on high-impact studies that lead future technological directions and solve complex problems within the machine learning landscape.

美团技术团队•July 13, 2026

Read AnalysisSource

Research Breakthrough

LongCat Open Sources VitaBench 2.0: A New Benchmark for Long-Term Dynamic AI Agent Evaluation

The Meituan technology team, under the LongCat initiative, has officially released VitaBench 2.0, a pioneering open-source benchmark designed for long-term dynamic user modeling. This benchmark represents a significant shift in AI evaluation, focusing on real-life scenarios rather than static tasks. VitaBench 2.0 is specifically engineered to systematically assess the performance of Large Language Models (LLMs) in two critical areas: personalization and proactivity. By simulating long-term, authentic, and evolving user interactions, the benchmark provides a standardized framework to measure how effectively AI agents can adapt to individual user needs over time. This release aims to address the complexities of sustained human-AI engagement in dynamic environments.

美团技术团队•July 13, 2026

Read AnalysisSource

Industry News

Meituan Technical Team Showcases Cutting-Edge AI Research in Search and Recommendation at Top Global Conferences

The Meituan Business R&D Platform's Search and Recommendation ASX (Agentic System X) team has recently highlighted its significant contributions to the field of Artificial Intelligence. Focusing on the development of Large Language Model (LLM)-based Agent systems, the team has achieved breakthroughs in LLM post-training, Agentic Reinforcement Learning, and multimodal understanding. These advancements have led to dozens of publications in prestigious international conferences, including ICLR, NeurIPS, CVPR, and AAAI. This article explores the team's strategic focus on building sophisticated Agentic systems and the implications of their research for the future of search and recommendation technologies. By selecting six key papers for in-depth interpretation, Meituan demonstrates its commitment to pushing the boundaries of AI application in complex service scenarios.

美团技术团队•July 13, 2026

Read AnalysisSource

Open Source

Meituan Open Sources Innovative AIGC Poster Generation System Featuring a Technical Closed Loop

The Meituan Intelligent Creation Team has officially unveiled and open-sourced its comprehensive technical system for AIGC poster generation. This framework is built around a unique "Generation-Editing-Evaluation" closed loop, designed to handle the complexities of industrial-grade visual content creation. By integrating these three core phases, Meituan has successfully implemented the system within its food delivery (Meituan Waimai) and Brand IP scenarios. The move to open-source this technology provides the global developer community with a structured approach to automated graphic design, emphasizing not just the creation of images, but the refinement and quality assessment necessary for commercial applications. This release marks a significant step in transitioning AIGC from experimental tools to scalable production pipelines.

美团技术团队•July 13, 2026

Read AnalysisSource

Industry News

Meituan Showcases AI Innovations at ACL 2026: Advancing Large Model Evaluation and Reasoning Optimization

Meituan's technical team has announced the acceptance of six research papers at the prestigious ACL 2026 conference, a leading international venue for computational linguistics and natural language processing. The selected works cover a diverse range of cutting-edge technical directions, including large model evaluation, complex process reasoning, and competition-level mathematical thinking optimization. Additionally, the research delves into reinforcement learning optimization and the emerging field of generative recommendation systems. These contributions highlight Meituan's strategic focus on building a new generation of generative AI paradigms, aiming to enhance both the theoretical capabilities and practical applications of large language models in complex, real-world scenarios.

美团技术团队•July 13, 2026

Read AnalysisSource

Industry News

Meituan LongCat Releases General 365: A New Benchmark Revealing Significant Gaps in AI Reasoning Capabilities

The Meituan LongCat team has officially launched General 365, a rigorous new benchmark designed to evaluate the reasoning capabilities of Large Language Models (LLMs). In an initial assessment of 26 mainstream models, the results underscore a significant performance gap in the industry. Gemini 3 Pro, currently recognized as one of the most powerful models, led the group but only achieved an accuracy rate of 62.8%. Most notably, the vast majority of the 26 models tested failed to reach the 60% passing threshold. This benchmark aims to establish a more demanding standard for AI evaluation, highlighting that complex logical reasoning remains a major challenge even for state-of-the-art systems.

美团技术团队•July 13, 2026

Read AnalysisSource

Research Breakthrough

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

The Meituan LongCat team has officially introduced and open-sourced WBench, a groundbreaking evaluation framework designed to measure the capabilities of interactive video world models. As the first systematic multi-round benchmark in this field, WBench serves as a diagnostic tool—likened to a 'CT scanner'—to identify the specific limitations AI models encounter when transitioning from passive observation to active interaction. By focusing on the boundaries of how AI simulates and responds to dynamic environments, WBench provides a structured approach to understanding the current state of world models. This initiative marks a significant step in the evolution of AI evaluation, moving beyond simple video generation to assess how models handle complex, multi-stage interactive scenarios.

美团技术团队•July 13, 2026

Read AnalysisSource

Research Breakthrough

Meituan Fulfillment AI Team Showcases LLM Agent Innovations and Research Breakthroughs at ACL 2026

The Meituan Fulfillment AI Algorithm Team has unveiled its latest advancements in Large Language Model (LLM) Agent technology at the ACL 2026 conference. Centered on empowering Meituan's fulfillment business, the team is developing a self-evolving Agent operating system. Their research spans critical areas including Continued Pre-training (CPT), Post-training, Agentic Reinforcement Learning (RL), and Multimodal Understanding. With dozens of high-quality papers published in prestigious international forums like ACL and EMNLP, Meituan is positioning itself at the forefront of AI-driven operational efficiency. This session highlights how the team integrates frontier AI research with practical fulfillment scenarios to create autonomous, self-improving systems that enhance service delivery and operational workflows.

美团技术团队•July 13, 2026

Read AnalysisSource

Open Source

Meituan Open Sources LongCat-Video-Avatar 1.5: Transitioning Digital Human Video Models to Commercial-Grade Applications

Meituan's technical team has officially announced the open-source release of LongCat-Video-Avatar 1.5, a significant evolution in digital human video modeling. Moving beyond experimental state-of-the-art (SOTA) benchmarks, this version is designed for robust commercial-grade applications. The update introduces comprehensive improvements in lip-sync accuracy, physical plausibility, and long-video stability. Additionally, it features enhanced support for multi-person interactions and optimized inference efficiency. By focusing on natural and high-quality output within complex commercial environments, LongCat-Video-Avatar 1.5 aims to bridge the gap between theoretical performance and real-world usability, effectively moving digital human technology from the 'rehearsal room' to the 'real stage' of diverse, large-scale applications.

美团技术团队•July 13, 2026

Read AnalysisSource

Product Launch

Claude Code Templates: New CLI Tool Emerges to Streamline AI Configuration and Monitoring

The developer community has seen the introduction of 'claude-code-templates,' a specialized Command Line Interface (CLI) tool designed to enhance the user experience for Claude Code. Developed by davila7 and gaining traction on GitHub Trending, this tool focuses on two critical aspects of the AI development workflow: configuration and monitoring. By providing a structured environment via an npm package, it allows developers to manage their Claude Code setups more efficiently. As AI-driven coding assistants become integral to software engineering, utility tools like claude-code-templates represent a growing ecosystem of 'meta-tools' aimed at optimizing local development environments and providing better oversight of AI interactions.

GitHub Trending•July 13, 2026

Read AnalysisSource

Open Source

OpenAI Releases Curated Repository of Codex Plugin Examples to Support AI Model Extensibility

OpenAI has officially launched a GitHub repository dedicated to providing curated examples of Codex plugins. This initiative is designed to offer developers a clear framework for extending the capabilities of the Codex model through a standardized plugin architecture. According to the repository documentation, each plugin is organized within a specific directory structure, requiring a mandatory configuration file located in the .codex-plugin/ directory. By providing these curated examples, OpenAI aims to demonstrate the practical application and integration of plugins within its ecosystem. This release serves as a foundational resource for developers seeking to build custom tools and enhancements for Codex, emphasizing a structured approach to AI software development and modular integration.

GitHub Trending•July 13, 2026

Read AnalysisSource

Product Launch

DesktopCommanderMCP: Empowering Claude with Terminal Control, File System Search, and Advanced Diff Editing Capabilities

DesktopCommanderMCP is a newly released Model Context Protocol (MCP) server developed by wonderwhy-er, specifically designed to enhance the capabilities of the Claude AI model. This tool bridges the gap between the AI and the local desktop environment by providing three core functionalities: comprehensive terminal control, efficient file system searching, and precise diff-based file editing. By integrating these features, DesktopCommanderMCP allows Claude to not only search and manage files but also execute terminal commands directly, transforming the AI into a more proactive assistant for developers and power users. This release highlights the growing ecosystem of MCP-based tools aimed at expanding AI's operational reach within local computing environments, offering a more integrated experience for users looking to automate complex desktop tasks through natural language interfaces.

GitHub Trending•July 13, 2026

Read AnalysisSource

Open Source

Google Labs Unveils Stitch-Skills: A Standardized Library for AI Agent Interoperability

Google Labs has introduced 'stitch-skills,' a specialized repository designed to enhance the capabilities of Stitch MCP (Model Context Protocol) servers. This library provides a collection of Agent Skills that strictly adhere to the Agent Skills open standard, ensuring seamless integration across a wide array of modern AI programming agents. By supporting platforms such as Gemini CLI, Claude Code, Cursor, and Antigravity, stitch-skills aims to bridge the gap between AI models and functional tool execution. The project represents a significant move toward standardizing how AI agents interact with external environments, providing developers with a consistent framework for building and deploying skills that work across different AI ecosystems without requiring platform-specific modifications.

GitHub Trending•July 13, 2026

Read AnalysisSource

Industry News

Samsung Targets 2029 for Yongin Chip Plant Operations Amid 16GW Power Demands

Samsung Electronics has established a 2029 operational target for its inaugural facility within the Yongin semiconductor cluster. This strategic timeline marks a significant milestone in the development of what is poised to become a central hub for global chip production. A primary focus of this development is the immense energy infrastructure required to support the site's operations. Projections indicate that the Yongin cluster will demand between 15 and 16 gigawatts of electricity once it reaches full capacity. This analysis explores the significance of the 2029 timeline and the implications of such substantial power requirements for the future of semiconductor manufacturing, emphasizing the critical link between industrial growth and energy stability.

Tech in Asia•July 12, 2026

Read AnalysisSource

Industry News

July 2026 Developer Trends: Exploring Modern Smalltalk VMs, AI-Assisted Game Design, and Cross-Cultural Philosophy

The July 2026 'Ask HN: What Are You Working On?' thread provides a unique window into the diverse projects currently occupying the developer community. From the intersection of literature and global philosophy to high-performance systems programming and AI-driven creative recovery, the contributions highlight three distinct areas of innovation. Key projects include 'Veritas,' a book cataloging universal cultural truths; a from-scratch Smalltalk Virtual Machine optimized for modern multicore hardware and remote web-based development; and a narrative-driven 2D strategy game being developed with the assistance of Claude Code. These initiatives reflect a broader industry trend toward leveraging generative AI for solo project execution, the modernization of legacy programming environments for contemporary hardware, and the use of technology to synthesize global human knowledge.

Hacker News•July 12, 2026

Read AnalysisSource

Industry News

Lorde Criticizes AI Smartglasses During Madrid Performance, Calling the Technology 'Not Sexy'

During her set at the Real Cool Festival in Madrid, acclaimed singer-songwriter Lorde voiced her disapproval of AI-integrated eyewear. Addressing the audience directly, she described the emerging technology as "not sexy," signaling a cultural critique of wearable AI. While Lorde did not explicitly name a brand, her comments were made at an event sponsored by Ray-Ban, which is currently in a high-profile partnership with Meta for their AI smartglasses line. This incident highlights a potential friction point between tech companies' push for mainstream AI adoption and the aesthetic values of prominent cultural figures. The critique focuses on the social and visual appeal of the devices rather than their technical capabilities.

The Verge•July 12, 2026

Read AnalysisSource

Industry News

Claude Code vs. OpenCode Token Efficiency Analysis: Why Claude Code Uses 33,000 Tokens Before Your First Prompt

A technical comparison between Claude Code and OpenCode reveals a significant disparity in token consumption and cache efficiency. The study found that Claude Code initiates sessions with approximately 33,000 tokens of system prompts, tool schemas, and scaffolding—nearly five times the 7,000 tokens used by OpenCode for identical tasks. While newer models like Claude Fable 5 narrow this gap to a 3.3x multiple, Claude Code remains substantially more "token-hungry." Furthermore, the analysis highlights severe cache inefficiencies in Claude Code, which rewrites up to 54x more cache tokens than OpenCode, leading to higher operational costs. With production configurations and subagents, token usage can balloon from 121,000 to over 513,000, raising critical questions about the overhead of agentic AI frameworks in regulated and high-scale industries.

Hacker News•July 12, 2026

Read AnalysisSource

Industry News

How Apple’s Cancelled Self-Driving Car Project Paved the Way for Modern High-Performance AI Silicon

Apple's ambitious self-driving car initiative, often referred to as Project Titan, may have failed to produce a commercial vehicle, but its technological legacy lives on through the company's advanced AI chips. Early in the project's development, Apple identified a critical need for massive on-device AI processing power to handle autonomous driving tasks. While the specific processor intended for the vehicle was never completed, the research and development efforts laid the groundwork for the powerful Apple Silicon seen in today's devices. According to reports from Mark Gurman, this pivot toward high-performance neural processing has become a cornerstone of Apple's hardware strategy, transforming a failed automotive venture into a significant advantage for its current AI capabilities.

The Verge•July 12, 2026

Read AnalysisSource