AI Daily: Tencent Releases Open-Source Flagship Model Hy4preview; Midjourney V8.2 Launches Image Editing Model; Google Gemini Omni 1.1 Flash Model Debuts

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/en

1. Tencent Hunyuan releases open-source flagship model Hy4preview with 770B parameters and 1M context length

Tencent Hunyuan released the open-source flagship model Hy4preview, which has a total of 770B parameters and 1M context length, targeting real productivity scenarios. The model has been launched on multiple platforms and has made significant progress in software engineering, office scenarios, game development, and scientific research. Internal blind test data shows its performance is better than other models, and through an open-source strategy, it is driving a shift in industry trends.

AiBase Summary:

🧠 Hy4preview has made significant progress in software engineering, office scenarios, game development, and scientific research.

📊 In internal blind tests, Hy4preview scored higher than GLM-5.3 and Kimi K3.

🚀 Tencent is driving the competition in large models from parameter contests to practical productivity verification through open-source strategies.

2. Google Launches New Gemini Omni 1.1 Flash Video Model with 4K Resolution Output

Google has launched a new Gemini Omni 1.1 Flash video generation AI model, achieving breakthroughs in high-definition and long-duration video generation in the multi-modal field. Its features include scene expansion, camera movement control, and the introduction of reference video clips, enhancing the flexibility and professionalism of video generation. At the same time, the model offers multiple resolution options and adopts a tiered pricing strategy, lowering the entry barrier.

AiBase Summary:

🎥 Scene expansion allows users to seamlessly continue generating subsequent frames at the end of an existing video, with each expansion step lasting 10 seconds, allowing up to 40 seconds of continuous video.

📽 Camera movement control enables smooth transitions and professional shots by specifying the start and end frames of a shot.

📹 The new model supports introducing reference video clips up to 3 seconds long, ensuring accurate contextual consistency and character continuity when building new scenes.

3. Midjourney V8.2 Launches Image Editing Model with Full Function Support for Instruction Fine-tuning and Canvas Expansion

The release of Midjourney V8.2 image editing model marks a significant upgrade in the AI painting field. The model provides creators with a more precise and controllable visual creation experience through features such as instruction fine-tuning, multi-image reference, and local re-drawing.

AiBase Summary:

🎨 Instruction editing capability allows users to modify existing images through text instructions.

🖼️ Multi-image reference generation mechanism, supporting up to 4 images as references at once.

🖌️ Local redrawing and canvas expansion functions make adjusting image details and extending dimensions easy and simple.

Details link: https://updates.midjourney.com/edit-model-for-v8/

4. Ant Group Collaborates with CICC to Launch the First Financial Enhancement Large Model Ling-3.0-flash-Fin

Ant Group and CICC have jointly launched the financial enhancement large model Ling-3.0-flash-Fin, which significantly enhances investment research capabilities after training on specialized financial corpus and performs outstandingly in core capabilities such as information retrieval. At the same time, both parties also launched the FinFIRST financial search benchmark and plan to open-source it to promote industry ecosystem development.

AiBase Summary:

🧠 The financial enhancement large model Ling-3.0-flash-Fin strengthens core investment research capabilities through specialized financial corpus training.

🔍 The model focuses on four core capabilities including information retrieval, showing outstanding performance and improved general capabilities.

🌐 Open-source plans promote industry ecosystem development, offering free API calls and open model weights.

5. Alibaba Launches Qoder: Upgrading AI Programming Tool into an Intelligent Agent Workbench for Everyone

Alibaba has launched a new Qoder, upgrading from an AI programming tool to an intelligent agent workbench for everyone, improving work efficiency and user experience.

AiBase Summary:

🧠 The new version of Qoder shifts the focus from "code engineering" to "intelligent agent tasks," improving work efficiency.

⚙️ Technically supports a self-contained loop of "read—modify—verify—iterate," with cross-file editing capabilities.

🌐 The product provides dual access points for programming mode and general mode, enabling non-technical users to participate in development.

6. Google Upgrades AI Search Mode: Supports Flight Tracking and Hotel Booking on Over 300 Platforms

Google has made a major functional upgrade to its conversational search experience "AI mode," adding features such as flight price tracking, hotel booking assistance, and viewing of points/mileage prices, marking its acceleration in turning AI mode into a full-process AI travel agent.

AiBase Summary:

✈️ Supports integrating real-time prices and recommendations from over 300 airlines and travel websites.

🏨 In hotel scenarios, it can recommend accommodations based on user preferences, including reviews, and seamlessly redirect to partner interfaces for payment.

✈️ Global users can query specific amounts for airline miles or hotel point exchanges through natural language.

7. XPeng's Second-Generation VLA Large Model Receives Its First Major Upgrade, With Safety Capabilities Increased by 20 Times

XPeng's second-generation VLA large model received its first major upgrade, with the core being enabling AI to truly understand time, achieving a leap from static 3D space to dynamic 4D spatiotemporal. At the same time, the model's inference efficiency was improved, introducing the master agent of the entire vehicle, and bringing the L4-level experience of Robotaxi down to mass-produced vehicles.

AiBase Summary:

🧠 Introduced the Infini-VLA long-sequence architecture, enhancing the model's understanding of time.

🚀 Improved model inference efficiency, achieving parallel processing of "watching, thinking, and acting."

🤖 Launched the Master Agent for the first time, realizing the cockpit integration of VLA and VLM.

8. Anthropic Launches MHS Hardware Standard, Entering the Physical AI Arena for the First Time

Anthropic released the Model Hardware Standard (MHS), marking its first public entry into embodied intelligence and physical AI fields. MHS is a unified control and communication standard that enables AI agents to control various devices, improving experimental efficiency. This technology shows strong speed-up capabilities in fields such as biopharmaceuticals, quantum computing, and advanced manufacturing.

AiBase Summary:

🧠 MHS is a unified control and communication standard that allows AI agents to control various devices.

🧪 In the biopharmaceutical field, MHS significantly shortens the development and operation cycle of experiments.

🚀 Anthropic's MHS standard extends the large model competition from dialogue to hands-on operation fields.