1. Chinese Version of Cursor! Tencent Launches AI Coding Assistant CodeBuddy
Tencent has launched the CodeBuddy 3.0 plugin, deeply integrated with the WeChat Mini Program development tool, supporting features like code completion and intelligent hints to enhance development efficiency.
[AiBase Summary:]
🌐 CodeBuddy 3.0 assists in WeChat Mini Program development, improving flexibility.
🛠️ Supports multiple development tools, integrates dual models for intelligent code completion and generation.
📱 Plugin design requires no change in development environment, compatible with game development and more functionalities.
Product link: https://copilot.tencent.com/
2. Only 20B Parameters! ByteDance Releases Seed1.5-VL Multimodal Model, Achieves 38 SOTA Results
ByteDance releases the Seed1.5-VL multimodal model, with only 20B parameters, achieving industry-leading performance in tasks such as video understanding, with low inference costs.
[AiBase Summary:]
📈 Seed1.5-VL has few parameters but strong performance, achieving SOTA on 38 tasks.
💰 Low inference cost, input costs $0.003 per 1,000 tokens, output costs $0.009.
🔗 Open API on Volcano Engine, developers can quickly call to build AI visual assistants, etc.
Product link: https://github.com/ByteDance-Seed/Seed1.5-VL
3. Free Access! Qwen "Deep Research" Functionality Launched on Tongyi Qwen
The Qwen team introduces the "Deep Research" intelligent assistant system, which can organize complex issues and generate research reports to improve research efficiency, now free for public experience.
[AiBase Summary:]
🔍 "Deep Research" can plan complex research tasks and integrate online information.
📝 Generate detailed reports with records and citations to improve efficiency.
🎉 Based on the Qwen model, suitable for researchers and general users alike.
4. Manus' Parent Company Butterfly Effect Issues Statement, Denies $1.5 Billion Financing Rumor
The Butterfly Effect company denies the rumor about a $100 million financing round at a valuation of $1.5 billion, stating that the news is "seriously inaccurate," and emphasizes its focus on product development.
[AiBase Summary:]
🌟 Butterfly Effect denies financing rumors and focuses on product development.
⚖️ The statement mentions that official channels will release financing information.
📜 Will seek legal action to protect its reputation.
5. Devv2.0 Officially Starts Private Beta: AI-Driven Developer Tools Receive Comprehensive Upgrade
Devv2.0 enters the Private Beta stage, transforming from an AI search engine into a developer productivity tool covering more stages of the development cycle.
[AiBase Summary:]
🚀 Devv2.0 transforms from search to generation and deployment, enhancing development efficiency.
🤖 Utilizes generative AI models to produce code and supports seamless deployment experiences.
👥 Invites community feedback during Private Beta to optimize the product.
6. Apple Unveils Revolutionary AI Model Matrix3D: Easily Converts Three Photos Into 3D Scenes
Apple unveils the Matrix3D model, generating 3D scenes from a small number of 2D photos, simplifying the modeling process and promoting the development of AI technology.
[AiBase Summary:]
📸 Matrix3D generates high-quality 3D reconstructions from three photos, simplifying operations.
💡 Integrates multiple processing steps, increasing efficiency and reducing human error.
🌐 Brings potential to immersive technology applications and promotes the development of the metaverse.
Product link: https://machinelearning.apple.com/research/large-photogrammetry-model
7. Anthropic's New AI Model Claude Neptune Set to Debut, Competing with Market Giants
Summary: Anthropic’s new model, Claude Neptune, enters internal safety testing and will compete with GPT-5 and others, focusing on user safety and privacy.
[AiBase Summary:]
🛡️ Claude Neptune undergoes internal safety tests, focusing on preventing jailbreak attacks.
📅 Expected to launch in late May or early June, strengthening multi-modal and agent functions.
🌟 Designed with emphasis on user safety and privacy, improving AI system stability.
8. Tsinghua University and Mianbi Intelligence Release End-Side GUI Agent: AgentCPM-GUI
Tsinghua University and Mianbi Intelligence introduce AgentCPM-GUI, based on the MiniCPM-V model, supporting both Chinese and English operations, covering over 30 mainstream Chinese apps.
[AiBase Summary:]
📱 AgentCPM-GUI takes mobile screen images as input, accurately operating app interface elements.
🧠 Enhances planning reasoning capabilities through RFT technology, improving task execution success rates.
🌟 Lightweight, high-performance model, promoting the popularization of end-side AI applications.
9. Reversing Fate! Flow-GRPO Transforms Image Generation Models into "Gods"
Flow-GRPO enhances image generation model performance through "ODE to SDE conversion" and "denoising reduction strategies," showing excellent performance in multiple tasks.
[AiBase Summary:]
🎨 Flow-GRPO transforms image generation models from "Bronze" to "King," improving generation quality.
🔍 Significantly improves model accuracy in tasks such as composite image generation.
🚀 Expected to expand into video generation in the future, facing new challenges.
Product link: https://github.com/yifan123/flow_grpo
10. Google Search Is About to Launch "AI Mode" to Replace the Classic "I'm Feeling Lucky" Button
Google is testing a new "AI Mode" button, planning to replace the "I'm Feeling Lucky" button to provide a smarter search experience. This feature is still in the experimental stage and only available to some users.
[AiBase Summary:]
🌟 Google is testing the "AI Mode" button, potentially replacing the "I'm Feeling Lucky" function.
🤖 The new feature aims to enhance search intelligence and personalization experiences.
🧪 Currently in the experimental phase, limited to certain users for testing.
11. Stepfun-X Opens Source 3D Large Model Step1X-3D
Step1X-3D is a 3D large model with 4.8B parameters, capable of generating high-fidelity and controllable 3D content, providing strong technical support for creation.
[AiBase Summary:]
1️⃣ Data-driven and algorithm协同optimization, solving data scarcity and quality issues.
2️⃣ Uses a two-stage architecture, ensuring reliable and realistic generated geometry structures.
3️⃣ Improves controllability and usability, supporting intuitive control of various attributes.