Welcome to the 【AI Daily】 column! This is your guide to exploring the world of artificial intelligence every day. Here we present the latest developments in the AI field daily, focusing on developers and helping you gain insights into technological trends and innovative AI product applications.
Fresh AI products click to learn more:https://top.aibase.com/
1. Qwen3 series models go viral globally: Downloads exceed 12.5 million, with over 130,000 derived models
One month after the release of the Qwen3 series models, global downloads surpassed 12.5 million within a month, performing exceptionally well on multiple mainstream AI open-source platforms, particularly with over 130,000 derived models on Hugging Face, ranking first globally.
【AiBase Summary:】
🚀 Open-source for just one month, global cumulative downloads exceed 12.5 million, showcasing strong appeal.
🌐 Offers multiple version choices, each exceeding one million downloads, covering a wide range of needs.
🌟 Derived model quantity exceeds 130,000, ranking first globally on Hugging Face, reflecting high innovation vitality.
2. Dora3.0 smart reference feature fully launched! One-click generation of film-level posters, AI design enters the "zero threshold" era!
The AI creation platform Dora AI from ByteDance has updated its smart reference feature, significantly lowering the design threshold so that ordinary users can easily create professional-level posters.
【AiBase Summary:】
✨ Strong Chinese language understanding capability, film-level generation effect, disrupting traditional design processes.
🌟 Supports one-click generation of design works that meet specific styles, covering various scenario applications.
💯 Accurate detail retention, low cost, high efficiency, suitable for users at all levels to quickly realize creativity.
3. Zhipu AI officially launches enterprise-level super assistant Agent CoCo
Today, Zhipu AI releases the enterprise-level super assistant Agent CoCo, with the core concept of 'understanding you and the company, being capable and deliverable', enhancing enterprise work efficiency.
【AiBase Summary:】
🌟 CoCo focuses on delivering results, assisting throughout the workflow to ensure maximized task outcomes.
💼 Introduces unique memory mechanisms, providing personalized services, actively tracking industry dynamics.
🔗 Can be seamlessly integrated into enterprise systems, merging existing resources to create exclusive intelligent assistants.
Details link: https://aiworker.aminer.cn/ai_worker/verification?utm_source=zhipuai_social&utm_medium=wechat&utm_campaign=p250609
4. Baidu launches large model for financial industry, intelligent body becomes new focus of AI competition
At the 2025 Intelligent Economy Forum, Baidu Cloud Intelligence released the Qianfan Hu Jin large model, specifically designed for the financial industry, aiming to provide more precise and efficient AI solutions. Shen Dou emphasized the importance of industry-specific large model construction and showcased Baidu's innovative achievements in intelligent bodies.
【AiBase Summary:】
📊 Baidu Cloud Intelligence launched the Qianfan Hu Jin large model, focusing on the financial sector, meeting the industry's high requirements for accuracy and real-time performance.
💼 Baidu has collaborated with 65% of central enterprises, proving that its cloud technology is widely recognized by the market.
🤖 Intelligent body becomes the new focus of AI competition, with Baidu enabling enterprise digital transformation through lightweight customization.
5. Xiaohongshu releases first open-source large model dots.llm1: 11.2 trillion non-synthetic data enhances Chinese performance
Xiaohongshu has released its first large-scale model dots.llm1, an expert hybrid model with 142 billion parameters, using 11.2 trillion non-synthetic high-quality data, performing excellently in Chinese tests.
【AiBase Summary:】
🌟 dots.llm1 uses an expert hybrid structure with 142 billion parameters, significantly reducing training and inference costs.
📊 Uses 11.2 trillion non-synthetic data, achieving an average score of 91.3 in Chinese tests, surpassing several competitors.
🔍 Introduces a rigorous data processing pipeline to ensure the effectiveness and reliability of high-quality training data.
Details link: https://huggingface.co/rednote-hilab/dots.llm1.base/tree/main
6. Robot arms can also “integrate large models”! Hugging Face LeRobot is open-sourced, drastically reducing the threshold for AI robot development!
Hugging Face’s LeRobot project provides an efficient and user-friendly AI development platform for robots by integrating advanced algorithms and development toolchains, significantly reducing hardware adaptation costs and technical barriers.
【AiBase Summary:】
Unified interface adapts to multiple hardware devices, reducing developers' hardware adaptation costs.
Pre-trained models built-in, supporting quick loading of state-of-the-art robot control models.
Intelligent evaluation and efficient training functions accelerate development processes and improve model reuse efficiency.
Details link: https://github.com/huggingface/lerobot
7. ChatGPT voice function upgrade, real-time translation dialogues become more natural and fluent
OpenAI has comprehensively upgraded the voice function of ChatGPT, including natural and fluent speech expression and newly added real-time translation functionality, but there are still issues with audio quality and "hallucinations."
【AiBase Summary:】
🌟 Speech becomes more natural and fluent, with richer emotional expressions.
🌍 Real-time translation function added, supporting multilingual dialogues.
⚠️ There are issues with audio quality fluctuations and generating strange sounds.
8. Google Gemini application monthly downloads exceed ChatGPT, but user activity remains insufficient
Since the end of April 2025, Google’s Gemini application has exceeded ChatGPT in global Android downloads, with weekly installations reaching over 6 million, but user activity is only 4.9%, far below ChatGPT's 42.52%. Despite significant download growth, Gemini faces the challenge of improving daily user engagement.
【AiBase Summary:】
🌟 Gemini application downloads reach 6 million per week, surpassing ChatGPT.
📉 ChatGPT’s downloads have dropped to 3 million per week, but user activity remains as high as 42.52%.
🔄 Gemini needs to increase daily user activity to ensure long-term competitiveness in the market.
9. MonkeyOCR震撼登场: A 3B small model outperforms Gemini
As a lightweight document parsing model, MonkeyOCR performs excellently in English document parsing tasks with a parameter count of 3B, especially showing significant improvements in formula and table parsing. It is not only fast but also adopts an innovative 'structure-recognition-relation' triplet paradigm, bringing a new technical direction to the industry.
【AiBase Summary:】
Monkey 🐒 MonkeyOCR with 3B parameters outperforms Gemini2.5Pro and Qwen2.5-VL-72B in various document parsing tasks, especially improving formula parsing by 15.0%.
Lightning ⚡ MonkeyOCR parsing speed reaches 0.84 pages per second, far surpassing MinerU and Qwen2.5-VL-7B, suitable for enterprise-level rapid response needs.
Gear 🔧 The 'structure-recognition-relation' triplet paradigm improves parsing accuracy while reducing resource demands, offering flexible AI parsing solutions for enterprises.
Details link: https://arxiv.org/abs/2506.05218
10. Google Veo 3 FAST/TURBO mode online! Five times the cost-effectiveness, AI video generation enters the "ultra-speed" era!
Google has launched the new FAST/TURBO mode for Veo3, greatly reducing video generation costs and increasing efficiency, while supporting video output with native audio, providing more possibilities for content creators.
【AiBase Summary:】
FAST/TURBO mode offers five times the cost-effectiveness, significantly reducing production costs, suitable for frequent video production needs.
Supports native audio generation, achieving synchronized sound and image, greatly enhancing immersive experiences.
Combining rapid generation with high-quality details, it meets diversified needs from social media to professional fields.
11. Google AI Studio policy change: Gemini2.5Pro model free access gets "throttled"
Google will adjust its AI model usage policies, stopping free access to the Gemini2.5Pro series models, moving towards a system based on API keys. However, free users can still use the Gemini2.0 series models, albeit with limited capabilities.
【AiBase Summary:】
💎 Google officially announced the cessation of free calls to the Gemini2.5Pro series models, transitioning to API key authentication.
🚀 Current free users can still use the Gemini2.0 series models, but their performance does not match that of Gemini2.5Pro.
🌟 Developers need to weigh performance against cost, as future high-performance models may become fully commercialized.