Welcome to the 【AI Daily】 column! Here is your guide to exploring the world of artificial intelligence every day. Each day, we present hot content from the AI field, focusing on developers to help you gain insights into technical trends and understand innovative AI product applications.
Latest AI products click to learn more: https://top.aibase.com/
1. ByteDance's Volcano Engine officially releases DouBao Large Model 1.6 and DouBao Video Generation Model Seedance 1.0 Pro
At the FORCE Origin Power Conference, ByteDance launched DouBao Large Model 1.6 and Seedance 1.0 Pro video generation model, marking a comprehensive upgrade of its AI cloud-native full-stack services. DouBao Large Model 1.6 has significant improvements in functionality and performance, with more competitive pricing, while Seedance 1.0 Pro excels in the video generation field.
【AiBase Summary:】
🔥 DouBao Large Model 1.6 adopts a unified pricing model, significantly reducing usage costs.
🎥 Seedance 1.0 Pro video generation model leads in text-to-video and image-to-video generation fields.
🌟 Provides "Think-and-Search" and "DeepResearch" functions, greatly enhancing work efficiency.
2. OpenAI launches o3-pro AI model: Higher reliability and tool integration, but sacrifices speed
OpenAI released the latest o3-pro AI model, focusing on improving reliability and tool integration, but with slower response times and higher costs.
【AiBase Summary:】
🌟 o3-pro provides high reliability and precision, suitable for complex problem handling, but with slower response times.
💰 High usage cost may limit some developers' willingness to try it out.
💻 Supports multiple tool functionalities, such as web search, file analysis, visual input reasoning, etc., but does not support image generation yet.
3. Figma Officially Launches MCP, Design-to-Code Conversion at One Click, Revolutionizing Development Workflow
Figma introduced the Dev Mode MCP server in beta version, standardizing the protocol to directly pass design data to AI coding tools, significantly improving the efficiency and accuracy of design-to-code conversion.
【AiBase Summary:】
✨ The Figma Dev Mode MCP Beta supports real-time transfer of design data to AI coding tools, eliminating manual steps and errors.
🌐 MCP acts as a 'USB-C interface', improving code output fidelity and consistency, promoting deep integration between design and development.
🤝 Combining Figma Make with MCP provides a complete workflow from prototype design to production-level code, challenging industry competitors.
Details link: https://www.figma.com/blog/introducing-figmas-dev-mode-mcp-server/
4. New King of AI Image Generation! Krea 1 Released, Free Trial Unlocks Infinite Creativity
Krea AI launched its first image generation model Krea1, solving texture, contrast, and composition issues in traditional AI image generation, and supporting personalized customization, providing higher creative freedom for artistic creation.
【AiBase Summary:】
✨ Krea1 achieves highly realistic textures and diverse style expression through technological innovation, applicable to various creation needs.
🎨 Supports style references and customized training, allowing users to upload samples or datasets to generate images that meet specific aesthetic standards.
🌟 Offers a free trial version, lowering the technical threshold, making it easy for designers and general users to experience AI art creation.
5. ByteDance's DouBao daily token calls exceed 16.4 trillion, market share reaches 46.4%
At the 'ByteDance FORCE Origin Power Conference', Tan Dai disclosed significant growth data for the DouBao large model, emphasizing its widespread application in multiple fields. ByteDance remains at the forefront of domestic leadership in technology strength and market share, and will continue to deepen the application of AI technology in the future.
【AiBase Summary:】
🔥 As of the end of May 2025, DouBao Large Model's average daily token call volume reached 16.4 trillion, an increase of 137 times compared to the initial stage.
📊 ByteDance occupies 46.4% of the domestic large model market share, ranking first.
🚀 It will further increase R&D investment, promoting the wider integration of AI capabilities into enterprise operations.
6. French AI Lab Mistral Releases New Inference Model Magistral, Small Version Now Available for Download
Mistral released its first inference model series Magistral, including Small and Medium versions available on different platforms. Magistral has advantages in multi-language support and answer speed, but its performance does not surpass competitors in certain benchmark tests.
【AiBase Summary:】
🚀 Magistral releases inference model series, Small version now open for download, Medium version in preview stage.
🌍 Supports multiple languages, fast answer speed, applicable to various enterprise scenarios.
📊 Magistral's performance falls short of competitors in some benchmarks, but has unique advantages.
Details link: https://mistral.ai/news/magistral
7. Apple Introduces ChatGPT into New System, Garden of Pictures Supports Ghibli Style AI Image Generation
Apple upgraded the Garden of Pictures function in iOS26, iPadOS26, and macOS26 systems, integrating ChatGPT's image generation capability, adding various art styles such as Ghibli anime style, providing more diversified creation options. Users can generate images through text descriptions or photos, but the generation speed is slow and there are content restrictions; free users have limited generation opportunities.
【AiBase Summary:】
✨ New options such as Ghibli anime style add more diversity to creations.
🎨 Generate images through text or photos, AI assists art creation.
💰 Limited generation opportunities for free users, upgrade to premium accounts for more chances.
8. OpenAI Drops a Bombshell! o3 Price Drops by 80%, o3-Pro Strongly Launched
OpenAI drastically reduced the price of the o3 model and launched the high-performance o3-pro model, significantly lowering developers' usage costs, triggering intense competition in the AI market.
【AiBase Summary:】
The price of the o3 model has dropped by 80%, now comparable to or even lower than GPT-4o, significantly reducing developer costs.
The newly launched o3-pro model targets high reliability requirements, with reasonable pricing, already available to ChatGPT Pro and Team users.
The o3 model performs excellently in various benchmark tests, and after the price reduction, it will promote more innovative applications, while the o3-pro offers high-performance options for enterprise users.
Details link: https://platform.openai.com/docs/models/o3
9. Hugging Face Publishes Open Weight Model Contribution List: Chinese Teams Qwen and DeepSeek Enter TOP15
The globally renowned open-source AI platform Hugging Face published an open weight model contribution list, with Chinese teams Qwen and DeepSeek successfully entering the top 15, demonstrating China's technological strength and influence in the global open-source AI domain.
【AiBase Summary:】
The Qwen team supported by Alibaba Cloud ranked among the top open-source large language models due to its outstanding performance in instruction-following and code generation tasks with the Qwen3 series models.
DeepSeek is renowned for its high-performance and low-cost R1 series models and shows competitive advantages in specific fields, with its lightweight version defeating Google Gemini2.5 Flash in math tests.
Qwen plans to explore multimodal technologies, while DeepSeek will launch the R2 model, continuously driving AI innovation and gaining global recognition.
Details link: https://huggingface.co/spaces/cfahlgren1/model-release-heatmap
10. Alibaba Opens Source MaskSearch! AI Learns Active Search + Multi-step Reasoning, Precise Solutions to Complex Problems
Alibaba Tongyi Lab opens sourced the new pre-training framework MaskSearch, which significantly enhances AI performance in solving complex problems through retrieval-enhanced masked prediction (RAMP) and reinforcement learning techniques.
【AiBase Summary:】
✨ MaskSearch enables AI to actively search and perform multi-step reasoning through the RAMP mechanism, improving performance in open-domain question answering and complex reasoning tasks.
🌟 DAPO algorithm combines format rewards and answer rewards to optimize the accuracy and logic of AI-generated answers, enabling small models to perform媲美大规模 models.
🌐 MaskSearch is completely open source, supporting various models, promoting the wide application of intelligent search and reasoning technologies in education, healthcare, and other fields.
Details link: https://github.com/Alibaba-NLP/MaskSearch
