AI News on June 22, 2026

Meituan Data Platform Unveils New BI Architecture Centered on Metrics Platform and Enhanced Computing Engines
Industry News

Meituan Data Platform Unveils New BI Architecture Centered on Metrics Platform and Enhanced Computing Engines

Meituan's technical team has introduced a transformative Business Intelligence (BI) architecture. By shifting the focus to a centralized metrics platform, the company addresses critical bottlenecks in traditional BI workflows. The new system leverages automatic semantics and enhanced computing to eliminate data caliber confusion—a common issue where different users derive different results from the same data—and to drastically improve query performance. This evolution represents a significant step in Meituan's data strategy, moving away from fragmented, personalized datasets toward a unified, high-performance analytical environment that ensures data integrity and operational efficiency across the enterprise. The practice highlights the importance of semantic consistency and computational optimization in modern data-driven decision-making processes.

美团技术团队
Meituan Showcases AI Innovations at ACL 2026: From Model Evaluation to Reasoning Optimization and Generative Paradigms
Industry News

Meituan Showcases AI Innovations at ACL 2026: From Model Evaluation to Reasoning Optimization and Generative Paradigms

Meituan's technical team has announced the acceptance of six research papers at ACL 2026, a premier international conference in computational linguistics and natural language processing. The papers cover a broad spectrum of cutting-edge AI fields, including large model evaluation, complex process reasoning, and competition-level mathematical thinking optimization. Additionally, the research explores advancements in reinforcement learning and generative recommendation systems. These contributions signify Meituan's strategic focus on building a new paradigm for generative AI, aiming to enhance the logical depth and practical utility of language models. By addressing both theoretical benchmarks and real-world application challenges, Meituan continues to position itself at the forefront of NLP research, contributing to the evolution of how AI systems reason, learn, and interact with users in complex environments.

美团技术团队
Meituan Open-Sources LongCat-Video-Avatar 1.5: A Commercial-Grade Leap for Digital Human Video Generation
Open Source

Meituan Open-Sources LongCat-Video-Avatar 1.5: A Commercial-Grade Leap for Digital Human Video Generation

The Meituan technical team has officially announced the open-source release of LongCat-Video-Avatar 1.5, a significant upgrade that transitions digital human technology from experimental state-of-the-art (SOTA) models to robust, commercial-grade applications. This latest iteration delivers comprehensive improvements across several critical dimensions, including lip-sync precision, physical plausibility, and long-form video stability. Designed to meet the rigorous demands of complex commercial environments, the model also introduces support for multi-person interactions and enhanced inference efficiency. By ensuring natural and high-quality content output, LongCat-Video-Avatar 1.5 aims to move digital human generation from controlled simulations to diverse, real-world scenarios, offering a scalable solution for high-fidelity video production.

美团技术团队
Meituan LongCat Team Launches General 365: A New Benchmark Revealing Critical Gaps in AI Reasoning Capabilities
Industry News

Meituan LongCat Team Launches General 365: A New Benchmark Revealing Critical Gaps in AI Reasoning Capabilities

The Meituan LongCat team has officially released General 365, a rigorous new benchmark designed to evaluate the reasoning capabilities of modern artificial intelligence. In an initial assessment of 26 mainstream models, the results reveal a significant performance gap across the industry. Even Gemini 3 Pro, currently identified as the most powerful model in the test, achieved an accuracy rate of only 62.8%. Furthermore, the vast majority of the models tested failed to reach the 60% threshold, which is traditionally considered a passing grade. This release by Meituan's technical team establishes a new standard for measuring logical depth in AI and highlights the substantial room for improvement in complex reasoning tasks.

美团技术团队
Managing AI Coding with Agent Evaluation: Meituan's Practice in Refactoring 310,000 Lines of Code
Industry News

Managing AI Coding with Agent Evaluation: Meituan's Practice in Refactoring 310,000 Lines of Code

Meituan's technical team has introduced a groundbreaking approach to managing AI-assisted development, focusing on the refactoring of 310,000 lines of code. As AI now generates over 90% of code in certain environments, the primary challenge has shifted from production speed to the management of AI's output quality. The team argues that without unified standards, AI can exponentially increase technical debt and system chaos. To combat this, Meituan implemented an 'Agent evaluation' mindset, utilizing four key pillars: technical debt sorting, rule construction, a standardized Refactoring SOP, and a Pre-PR (Pull Request) mechanism. This strategy successfully transitions code refactoring from a high-cost, specialized project into a sustainable, daily iterative process, ensuring long-term system stability in the era of AI-dominated coding.

美团技术团队
LARYBench Released: Redefining Embodied AI Action Representation Through Large-Scale Human Video Learning
Research Breakthrough

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

The Meituan Technical Team has officially released LARYBench (Latent Action Representation Yielding Benchmark), a systematic evaluation framework designed to measure general latent action representations derived from large-scale visual data. This benchmark marks a significant milestone in embodied intelligence, often compared to the 'ImageNet' moment for action representation. The research findings reveal a paradigm shift: general-purpose vision models significantly outperform specialized embodied expert models in both action generalization and control precision. Crucially, the study demonstrates that embodied action representations can spontaneously emerge from large-scale human video data, providing a new pathway for developing more capable and generalized robotic systems without relying solely on specialized datasets.

美团技术团队
Meituan LongCat-AudioDiT: Breaking Zero-Shot TTS Limits via Direct Waveform Latent Space Diffusion
Research Breakthrough

Meituan LongCat-AudioDiT: Breaking Zero-Shot TTS Limits via Direct Waveform Latent Space Diffusion

The Meituan LongCat team has officially released LongCat-AudioDiT, a groundbreaking model designed to push the boundaries of zero-shot Text-to-Speech (TTS) and voice cloning. By fundamentally reimagining the audio synthesis pipeline, the team has moved away from traditional intermediate representations such as Mel-spectrograms. Instead, LongCat-AudioDiT operates directly within the waveform latent space using a diffusion-based architecture. This strategic shift is designed to eliminate the cascade errors typically caused by multi-stage data conversions. By allowing the AI to learn the inherent patterns of sound directly, the model aims to achieve a higher level of fidelity and accuracy in voice cloning, providing a more streamlined and robust solution for high-quality audio generation.

美团技术团队
Meituan Open-Sources LongCat-Flash-Prover: Advancing AI from Numerical Calculation to Rigorous Mathematical Theorem Proving
Open Source

Meituan Open-Sources LongCat-Flash-Prover: Advancing AI from Numerical Calculation to Rigorous Mathematical Theorem Proving

The Meituan Technical Team has announced the open-sourcing of LongCat-Flash-Prover, a specialized model designed to tackle the complexities of mathematical formalization and theorem proving. While traditional AI models often focus on achieving correct numerical outputs, LongCat-Flash-Prover addresses the more demanding requirement of maintaining strict logical chains. By focusing on formalization, the model seeks to eliminate the risks associated with natural language ambiguity, which can cause mathematical proofs to fail. This release marks a significant shift in AI development, moving from models that merely "guess" answers to systems capable of providing rigorous, verifiable mathematical proofs through structured reasoning.

美团技术团队
Meituan Open Sources LongCat-Next: A Native Multimodal Model Designed for Physical World AI Interaction
Open Source

Meituan Open Sources LongCat-Next: A Native Multimodal Model Designed for Physical World AI Interaction

Meituan's technical team has officially announced the release and open-sourcing of LongCat-Next, a pioneering native multimodal model. This release marks a significant step in Meituan's exploration of "Physical AI," where vision and speech are integrated as native components rather than secondary inputs. By open-sourcing the core model alongside its discrete tokenizer, Meituan aims to provide the global developer community with the essential tools to build AI systems capable of perceiving, understanding, and interacting with the real world. The project emphasizes a shift toward AI that treats sensory data as a primary language, potentially transforming how machines navigate and function within physical environments. This strategic move highlights Meituan's commitment to fostering an open ecosystem for advanced multimodal research and practical AI applications.

美团技术团队
Meituan LongCat Team Unveils WBench: The First Systematic Multi-Round Benchmark for Interactive Video World Models
Research Breakthrough

Meituan LongCat Team Unveils 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 systematic multi-round evaluation benchmark designed specifically for interactive video world models. Positioned as a diagnostic 'CT scanner' for artificial intelligence, WBench is engineered to precisely identify the technical limitations and performance bottlenecks encountered by world models as they transition from passive observation to active interaction. By evaluating models across diverse scenarios—ranging from lunar environments to complex cybernetic cities—WBench provides a framework for measuring how AI navigates the boundaries of simulated reality. This open-source initiative aims to standardize the assessment of interactive capabilities, offering the research community a vital tool to refine how AI systems perceive, simulate, and respond to dynamic, multi-stage user interactions within virtual environments.

美团技术团队
Google Research Unveils TimesFM: A Pretrained Foundation Model for Advanced Time Series Forecasting
Research Breakthrough

Google Research Unveils TimesFM: A Pretrained Foundation Model for Advanced Time Series Forecasting

Google Research has introduced TimesFM (Time Series Foundation Model), a pioneering pretrained foundation model specifically engineered for time series forecasting. Moving beyond traditional task-specific models, TimesFM applies the foundation model paradigm—successful in NLP and computer vision—to the complexities of temporal data. Developed by the expert team at Google Research, this model is designed to provide a robust, pretrained base that can be adapted for various forecasting scenarios. By leveraging large-scale pretraining, TimesFM aims to capture universal temporal patterns, offering a new level of efficiency and accuracy for researchers and industries dealing with time-dependent data. The project, highlighted on platforms like GitHub, represents a significant step forward in making sophisticated predictive analytics more accessible and scalable across diverse domains.

GitHub Trending
Palmier Pro: A Specialized AI-Native Video Editing Solution Launched for macOS
Product Launch

Palmier Pro: A Specialized AI-Native Video Editing Solution Launched for macOS

Palmier Pro has emerged as a new contender in the creative software market, specifically designed as a video editor for the macOS platform with a foundational focus on artificial intelligence. Recently gaining traction on GitHub, the project distinguishes itself by being built from the ground up for AI workflows rather than simply integrating AI as an afterthought. While the initial release information is concise, it highlights a significant trend toward platform-specific, AI-centric creative tools. This analysis explores the implications of Palmier Pro's entry into the macOS ecosystem, its positioning as an AI-native application, and what its presence on GitHub Trending suggests about the current state of open-source and specialized video production software.

GitHub Trending
OpenMontage: The World's First Open-Source Agentic Video Production System Debuts on GitHub
Open Source

OpenMontage: The World's First Open-Source Agentic Video Production System Debuts on GitHub

OpenMontage has launched as a pioneering open-source project, marking the arrival of the world's first 'Agentic' video production system. Developed by creator calesthio, the system is designed to transform standard AI programming assistants into comprehensive video production studios. The framework is built upon a massive architecture consisting of 12 specialized pipelines, 52 integrated tools, and a library of over 500 distinct agent skills. By providing an open-source alternative for complex multimedia creation, OpenMontage enables AI agents to handle multi-step video generation tasks autonomously. This release represents a significant milestone in the evolution of AI-driven content creation, shifting the focus from simple generative models to integrated, tool-augmented agentic workflows.

GitHub Trending
High-Performance Code Intelligence: Exploring the codebase-memory-mcp Server for Efficient Knowledge Graph Indexing
Open Source

High-Performance Code Intelligence: Exploring the codebase-memory-mcp Server for Efficient Knowledge Graph Indexing

The emergence of codebase-memory-mcp, a high-performance Model Context Protocol (MCP) server developed by DeusData, marks a significant advancement in code intelligence. By indexing codebases into persistent knowledge graphs, the tool achieves millisecond-level processing per repository and sub-millisecond query speeds. Supporting 158 programming languages, it is designed to reduce AI token consumption by 99%, addressing one of the primary cost and context window constraints in modern AI-assisted development. As a single static binary with zero dependencies, it offers a streamlined solution for developers seeking to integrate deep codebase understanding into their AI workflows without the overhead of complex infrastructure.

GitHub Trending
Samsung and SK Hynix Profit Forecasts Surge Amid Global Memory Shortage and Server DRAM Prioritization
Industry News

Samsung and SK Hynix Profit Forecasts Surge Amid Global Memory Shortage and Server DRAM Prioritization

The semiconductor industry is witnessing a significant upward revision in financial expectations for South Korean tech giants Samsung and SK Hynix. According to recent reports from TrendForce, profit forecasts for these companies are surging, primarily driven by a persistent global memory shortage. The analysis indicates that suppliers are strategically shifting their production focus toward server DRAM. This move is motivated by the significantly higher profitability found in the server-grade segment compared to other memory products. As Samsung and SK Hynix prioritize these high-margin components, the market dynamics are shifting to favor enterprise-level infrastructure, resulting in a bullish outlook for the leading memory manufacturers despite broader supply constraints.

Tech in Asia
Nasdaq-Bound Arms Maker UVision Targets $4 Billion Valuation for HERO Loitering Munitions Portfolio
Industry News

Nasdaq-Bound Arms Maker UVision Targets $4 Billion Valuation for HERO Loitering Munitions Portfolio

UVision, a prominent arms manufacturer, is seeking a $4 billion valuation as it prepares for its debut on the Nasdaq exchange. The company is recognized for its HERO loitering munitions, which offer versatile deployment options including man-portable and vehicle-launched configurations. This strategic financial move underscores the company's positioning within the global defense sector and highlights the growing market interest in specialized loitering munitions technology. As the company moves toward its Nasdaq listing, the $4 billion target sets a significant milestone for the firm and the broader defense industry, reflecting the value placed on portable and vehicle-integrated munitions systems. The transition to a public listing suggests a strategic intent to scale operations and capitalize on the demand for advanced defense hardware.

Tech in Asia
Industry News

Apertus Launches Apertus Mini: 16 Open Foundation Models Advancing Sovereign AI Through Distillation and Quantization Techniques

Apertus has officially released Apertus Mini, a specialized collection of 16 small language models designed to advance the concept of Sovereign AI. This release serves as a technical demonstration of how open foundation models can be optimized for efficiency and performance. The core focus of the Apertus Mini suite is to showcase the practical application of distillation and quantization techniques in model development. By providing a diverse set of 16 models, Apertus aims to provide the industry with a clear roadmap for creating high-performance AI that remains accessible and transparent. This initiative aligns with the broader movement toward Sovereign AI, emphasizing the importance of open-source architectures that allow for localized control and reduced reliance on proprietary, black-box systems.

Hacker News
Recall: A Fully-Local Project Memory Tool for Claude Code to Save Tokens and Enhance Privacy
Product Launch

Recall: A Fully-Local Project Memory Tool for Claude Code to Save Tokens and Enhance Privacy

Recall is a newly introduced fully-local project memory tool designed to solve the "cold-start" problem for Claude Code users. By maintaining a local log of user sessions and condensing them into a compact summary, Recall eliminates the need for developers to re-explain their projects at the start of every new session. Unlike many memory tools that rely on external LLMs, Recall utilizes a classical Python summarizer that runs entirely on the user's machine. This approach ensures that sensitive data, including code and secrets, never leaves the local environment while significantly reducing token consumption. By resuming from a condensed context file of approximately 1–2K tokens, users can stretch their Claude subscription limits or lower their API costs. Recall is designed to be zero-friction, requiring no API keys or complex installations, and functions as a complementary addition to Claude Code's native capabilities.

Hacker News
Technical Tutorial

Mastering JSON-LD: A Comprehensive Guide to Enhancing Personal Websites with Structured Data

In a detailed exploration of modern web optimization, developer Ethan Hawksley explains the implementation and benefits of JSON-LD (JSON Linked Data) for personal websites. Based on approximately 100 hours of coding and extensive research, the analysis highlights how structured data serves as a vital tool for web crawlers to interpret site semantics. By integrating specific script tags and adhering to Schema.org standards, website owners can qualify for enhanced link previews and potentially improve their search engine rankings. The guide breaks down the fundamental components of a JSON-LD script, including the importance of MIME types, the role of the @context property, and the organizational structure of the @graph array, providing a technical roadmap for developers looking to polish their digital presence.

Hacker News