Welcome to the "AI Daily" section! This is your guide to exploring the world of artificial intelligence every day. Every day, we present you with the latest content in the AI field, focusing on developers to help you understand technology trends and learn about innovative AI product applications.

Fresh AI products Click to learn more: https://app.aibase.com/zh

1. Challenge OpenAI, SpaceXAI launches "Opus-level" large model Grok4.5

SpaceXAI has launched a new generation of large model Grok4.5, which has significant improvements in performance and efficiency, and has a competitive pricing strategy, aiming to challenge competitors such as OpenAI.

image.png

【AiBase Summary:】

🚀 Grok4.5 is positioned as a versatile "main force," capable of handling complex tasks such as code writing, application development, office document processing, and knowledge-based writing.

💡 Grok4.5 has significantly improved token efficiency in performance optimization, producing more efficient results with less computing power, thereby directly reducing inference costs.

💰 SpaceXAI offers a highly competitive pricing plan: $2 per million input tokens and $6 per output token, attracting cost-sensitive enterprises and developers.

2. New heights in AI creation: ByteDance releases Seedream 5.0 Pro, opening the era of interactive precise editing

ByteDance released Seedream 5.0 Pro, marking the transition of AI image creation from generation to design. The model has significantly improved in text-image matching, structural rationality, text rendering, and visual quality, and has broken through four core capabilities, including visualization of complex information, interactive precise editing, realistic image and portrait quality, and native multilingual input and generation.

image.png

【AiBase Summary:】

🧠 The ability to visualize complex information: Seedream 5.0 Pro can deeply analyze user intent, accurately converting complex data, concepts, and dense text into professional layout charts.

🖌️ Interactive precise editing: The model achieves highly controllable local editing through understanding of spatial position and regional semantics, lowering the threshold for designers to make micro-adjustments.

🌍 Native multilingual input and generation: Supports more than ten languages, automatically adapting to local layout rules and cultural contexts, accurately restoring localized features.

3. Major upgrade for OpenAI's voice assistant: GPT-Live full-duplex model launched, ChatGPT finally can listen and speak at the same time

OpenAI released a new ChatGPT voice experience based on the GPT-Live architecture, achieving a full-duplex interaction mode, improving the naturalness and smoothness of the voice assistant, and supporting complex task processing. At the same time, two versions were launched to meet the needs of different users.

image.png

【AiBase Summary:】

🧠 GPT-Live architecture enables full-duplex voice interaction, enhancing the naturalness and fluency of conversations.

🔄 The new voice version supports complex task processing, such as online search and deep reasoning.

👥 OpenAI launched two versions to meet the needs of different users, covering both free and paid users.

4. MiniMax plans to launch a new large model with 2.7 trillion parameters

Xiyu Technology is preparing to release a new large model, with parameter scale reaching 2.7 trillion, which will significantly enhance the model's capabilities in complex task processing and logical reasoning. The release of this model not only demonstrates Xiyu Technology's investment in fundamental model R&D but also establishes a technical barrier in the intense competition for computing power. The industry believes that this model will play a greater role in applications such as dialogue interaction and intelligent analysis, and the market is full of expectations for its performance.

【AiBase Summary:】

🧠 Parameter scale reaches 2.7 trillion, marking an improvement in intelligence level.

🚀 Xiyu Technology establishes a technical barrier in the competition for computing power.

💡 The model will improve response speed and accuracy in complex scenarios.

5. Ant Lingbo opensources LingBot-Video, the world's first video foundation model for embodied intelligence!

Ant Lingbo open-sourced LingBot-Video, the world's first video generation foundation model designed for embodied intelligence. The model performed well in the RBench benchmark test, demonstrating strong physical understanding and action consistency. Its innovative architecture and training method make it more suitable for robot operations and real-time interaction needs.

image.png

【AiBase Summary:】

🧠 LingBot-Video is the world's first video generation foundation model based on MoE architecture, specifically designed for embodied intelligence.

🧪 In the RBench benchmark test, LingBot-Video outperformed multiple mainstream models, showing stronger physical understanding and action consistency.

🚀 Through DiT + MoE architecture, embodied data, and a multidimensional reinforcement learning reward system, LingBot-Video achieves efficient reasoning combined with real physical laws.

6. College entrance exam admission starts: Qwen integrates EMS to provide real-time tracking and reminders for admission letters

The article introduces Qwen integrating China Post EMS, offering real-time tracking and reminder functions for admission letters, helping users better understand the delivery progress of their admission letters. It also mentions that Qwen provides personalized services during the college application process, such as generating application reports and calendar functions for applications.

image.png

【AiBase Summary:】

✅ Qwen integrates EMS, providing real-time tracking and reminder functions for admission letters.

📌 Qwen helps users clearly plan the steps of application through the application calendar function.

📈 As of July 8th, users have collected 23 million free application reports, with high usage of the application calendar.

7. Jieyu Star's first AI agent phone will be released, leading ahead of OpenAI

Jieyu Star, one of Shanghai's six AI startups, is rapidly rising by leveraging its unique end-to-end large model plus terminal hardware strategy. The company's upcoming AI agent phone not only leads in technology but also builds a strong ecosystem through deep collaboration with industry chain companies such as Huaqin Technology.

image.png

【AiBase Summary:】

📱 Jieyu Star will launch the world's first native AI agent phone, leading ahead of OpenAI's AI terminal plan.

🤝 Huaqin Technology provides deep collaboration for manufacturing, forming an integrated chain of large models and hardware.

🚀 Jieyu Star was founded by former Microsoft executives, raising over 20 billion yuan to aim for a Hong Kong IPO.

8. A regular camera can navigate on its own: Mistral releases 8B model Robostral Navigate, performance surpasses multi-camera solutions

Mistral released the AI model Robostral Navigate for robot navigation, which can achieve autonomous navigation in complex environments using just a regular RGB camera. The model outperforms multi-camera and depth sensor solutions, showcasing its potential in embodied intelligence.

image.png

【AiBase Summary:】

🤖 Robostral Navigate achieves autonomous navigation in complex environments using just a regular camera.

📊 In the R2R-CE benchmark test, the success rate reached 76.6%, exceeding multi-camera and depth sensor solutions.

🌐 The model is compatible with wheeled, legged, and flying robots, offering broad applicability.