Manus, an AI platform from Chinese startup Monica, has garnered significant attention despite using pre-existing models like Claude and Qwen. Claims of its superiority over tools like OpenAI's Deep Research have fueled excitement, growing its Discord community to over 138,000 members. However, early users report flaws, including factual inaccuracies and incomplete tasks. While in beta testing, Manus struggles with simple requests like food ordering or flight booking, highlighting a gap between h

An AI agent product called Manus, developed by Monica.im, recently gained popularity for assisting with daily and work tasks. However, its official account was suspended on March 7 due to potential links to cryptocurrency scams mentioned by third parties. Co-founder Ji Yichao clarified that Manus is not involved in any crypto or blockchain projects and warned of fraudulent entities using similar names. The team is working to resolve the suspension and taking legal action against imposters. Furth

OpenAI upgrades the macOS ChatGPT app, enabling direct code editing in IDEs like Xcode and VS Code. The new "Collaborate with Apps" feature allows users to interact with ChatGPT within these environments using shortcuts for coding assistance. An "Auto-Apply" mode lets ChatGPT make changes without extra approval. This update is rolling out to paid subscribers first, with free users and Windows support coming soon. OpenAI plans to expand this functionality to more Mac apps, positioning ChatGPT as

AMD launches Instella, a new open-source language model with 3 billion parameters. Built on an autoregressive transformer architecture, it features 36 decoder layers and 32 attention heads, handling up to 4096 tokens. Trained on AMD Instinct MI300X GPUs, Instella shows 8% better performance than similar models. Its instruction tuning enhances interactive tasks, making it ideal for query understanding and context-aware responses. AMD promotes transparency by releasing model weights, datasets, and

ChatGPT regained strong growth momentum in H2 2024, reaching 400 million weekly active users by February 2025. Key drivers include new models like GPT-4o and features such as advanced voice mode. Mobile usage grew consistently, with 175 million weekly users. Competitor DeepSeek captured 15% of ChatGPT's mobile base but remains second in engagement. The report also highlights market trends and rankings for top AI apps.

Alibaba's stock surged over 8% following the release of QwQ-32B, a 32.5-billion-parameter open-source model performing on par with DeepSeek-R1. Built on the Transformer architecture with rotary position encoding, it excels in reasoning tasks like code generation and external application execution. Trained via reinforcement learning in two phases focusing on coding/math skills and problem-solving, QwQ-32B outperformed R1 in three of five benchmarks. Alibaba plans to invest $53 billion in AI infra

Tavus, an AI research startup, launched groundbreaking models for human-like interactions. Their "Conversational Video Interface" enables natural AI perception and response. Key products include Phoenix-3 for realistic facial expressions, Raven-0 for scene interpretation, and Sparrow-0 for fluid dialogue management. These integrated models aim to create immersive experiences resembling real human interactions, with ongoing improvements planned.

Mistral launches Mistral OCR API for advanced document understanding. It extracts structured data from unstructured PDFs and images, supports multiple languages, integrates with LLMs, and processes content like handwritten notes, tables, and formulas. This tool aims to help organizations digitize 90% of business information that exists as unstructured data.

Anthropic updates its developer platform to improve AI implementation for businesses. The new Anthropic Console enables team collaboration on AI prompts and integrates the advanced Claude 3.7 Sonnet model. This addresses coordination challenges by offering shared prompt creation and better control features, reducing knowledge silos and enhancing cross-departmental cooperation.

Larry Page has launched a new AI company, Dynatomics, focusing on advanced large language models for optimized solutions. It plans to integrate AI with manufacturing by building a factory for AI-designed products. Led by Chris Anderson, former CTO of Kitty Hawk, the company aims to redefine smart production. This move marks Page's latest step into AI-driven innovation.

DuckDuckGo's AI-generated answers have exited the testing phase, expanding sources beyond Wikipedia to the entire web. The company plans to integrate its AI chatbot with search functionality, offering privacy-focused features like local chat storage and anonymous use. Users can choose from multiple AI models, and Duck.ai will soon support web searches, voice interaction, and image uploads. While remaining free, advanced models may be subscription-based. DuckDuckGo aims for seamless integration o

Tencent HunYuan launched an open-source image-to-video model that generates 5-second clips with sound effects from a single image and motion descriptions. It also supports lip-sync, motion-driven features, and diverse character/scene types like realistic or anime. With 13 billion parameters, the model extends HunYuan's text-to-video capabilities and includes weights, inference, and LoRA training code for customization. Available on Github and HuggingFace, it has gained over 8.9K stars and inspir

Manus, the world's first universal AI agent developed in China, transforms ideas into actions with reasoning and practical outcomes. It excels in GAIA benchmark tests, surpassing rivals like OpenAI. Currently in invite-only beta testing, Manus embodies the philosophy of integrating knowledge with action, aiming to extend human potential by equipping LLMs with tool utilization skills. The company plans further refinements post-launch.

Apple is introducing an AI-powered summary feature for App Store reviews in iOS 18.4 and iPadOS 18.4, currently in beta. Leveraging large language models, it condenses user reviews into brief summaries, updated weekly for apps with sufficient reviews. Initially launching in English in the US, it will expand globally within a year. Users can report issues via long-press, while developers use App Store Connect. Concerns exist about potential misuse by unethical developers and over-reliance on summ

Prime Video is testing AI-assisted dubbing for select movies and TV shows, offering English and Latin American Spanish options. The service combines AI with human experts to ensure quality and will initially cover 12 titles. This move follows other streaming platforms adopting similar tech, like Paramount+ and YouTube, as AI dubbing enhances content accessibility. Prime Video also offers other AI features such as X-Ray Recaps and Dialogue Boost.

A US judge dismissed Elon Musk's request to halt OpenAI's transition to a for-profit entity, citing failure to meet the high standard for a preliminary injunction. The trial will proceed expedited between September and November 2025. OpenAI's shift aims to attract capital and boost competitiveness in AI, with its non-profit division retaining shares but losing operational control. Musk, who co-founded OpenAI and later left, has proposed acquiring its non-profit entity for $97.4 billion, but CEO

Contextual AI launched its Grounded Language Model (GLM), scoring 88% factuality on the FACTS benchmark, surpassing Google, Anthropic, and OpenAI models. GLM is optimized for enterprise RAG applications, prioritizing precision in regulated industries where factual accuracy is critical over creative flexibility.

Alibaba's QwQ-32B, a 32-billion-parameter reasoning model, leverages reinforcement learning to excel in complex problem-solving. Open-sourced under Apache 2.0, it supports commercial and research use. Built on a 64-layer transformer architecture, it integrates structured self-questioning and multi-stage RL for enhanced math, coding, and general reasoning. Despite fewer parameters than competitors like DeepSeek-R1, QwQ-32B achieves competitive performance with lower memory requirements.

OpenAI announces GPT-4.5 availability for ChatGPT Plus subscribers, showing significant improvements over GPT-4o in various tasks. The rollout may take 1-3 days with initial server delays. GPT-4.5 offers more natural interactions, an expanded knowledge base, and higher emotional intelligence. Costs for GPT-4.5 are substantially higher, with input priced at $75 per million tokens and output at $150, compared to $2.5 and $10 for GPT-4o. OpenAI plans to launch GPT-5 by May and phase out the o3 mode

Google's AR beauty tool now lets US users virtually try on celebrity-inspired makeup looks and buy matching products. The update also includes virtual try-on for clothing like pants and skirts, enhancing personalized shopping with AI technology.