Microsoft Azure is now offering a free trial of Grok, which will run through early June.
Users can try out this powerful AI tool on Azure Foundry and GitHub.Grok, an advanced natural language processing model, can help developers and organizations improve productivity.
One reservation, many models. Foundry Models now offers multiple partner models with performance guaranteed by the same SLA as Azure OpenAI. Soon, users will be able to use multiple models from OpenAI, xAI, DeepSeek, Meta, Mistral AI, and Black Forest Labs through a unified, pre-configured throughput.
This integration not only improves developer flexibility, but also simplifies resource management and provides more efficient support for multi-model applications.
---
Microsoft has announced that the Azure AI Foundry Agent Service is now live and provides top-level support for LlamaIndex.
This new Agent Service is designed to help enterprise customers build many types of intelligent assistants, including:
- Customer service assistant to efficiently handle customer inquiries and reduce response time
- Process automation robots that can orchestrate complex business processes
- Multi-agent system, allowing specialized agents to collaborate on tasks
- Enterprise-grade solution that securely integrates with your data and tools
Azure AI Foundry Agent ServiceThe advantage lies in its strong scalability and flexibility. Enterprises can customize intelligent solutions according to their needs, thus improving operational efficiency and service quality. In addition, the support of LlamaIndex enables Agent Service to excel in natural language processing and better understand and respond to user needs.
Whether in the areas of customer service, process management or data analytics, theAzure AI Foundry Agent ServiceAll of them provide strong technical support for enterprises and help them realize intelligent transformation.
---
Meta recently released two important tools designed to accelerate molecular chemistry and materials science:Open Molecules 2025 (OMol25) respond in singing Universal Model for Atoms (UMA)The
OMol25 is a new dataset based on Density Functional Theory (DFT) covering high-precision quantum chemical calculations of biomolecules, metal complexes and electrolytes. It is one of the largest and most diverse datasets to date, providing a rich reference resource for researchers.
UMA, on the other hand, is a machine-learning based model of interatomic potentials. By training data from more than 30 billion atoms, UMA is able to provide more accurate predictions and enhance the understanding of molecular behavior. This will greatly facilitate the design and development of new materials.
- OMol25: Highly accurate DFT dataset for biomolecules, metal complexes and electrolytes
- UMA: Machine Learning Model Based on 30 Billion Atoms Improves Accuracy of Molecular Behavior Predictions
The release of these two tools will greatly enhance the efficiency of research in molecular and materials science, opening up new possibilities for future technological innovation.
Related papers, models and datasets can be downloaded from the following link: https://t.co/4kggBhnRKu
---
In the latest news, Google is launching a beta version of an automated programming assistant called Jules globally. The tool, designed to compete with existing programming aids such as Codex, is capable of completing code submissions (PRs) independently and offers five free tasks per day.
Jules' core function is to analyze project requirements and automatically generate execution plans by connecting to GitHub. This automated solution not only improves development efficiency, but also dramatically reduces human error. For developers, Jules significantly reduces the daily repetitive workload, allowing them to focus more on complex and innovative tasks.
Key features include:
- Automated Code Submission
- 5 free uses per day
- Seamless GitHub integration
- Intelligent generation of execution plans
With the introduction of Jules, development teams will be able to manage and maintain code bases more efficiently, resulting in faster software development cycles.
---
Recently, Bright Data launched a powerful MCP (Multi-Connection Protocol) server that enables AI agents to access, search, crawl, and interact with web pages without hassle.
This MCP server offers more than 30 powerful tools designed to address common blocking issues AI faces during web access. With these tools, AI can simulate the behavior of real users, thus avoiding being identified as an automated program by websites and getting blocked.
Highlighted features include::
- Highly scalable to support large-scale deployments
- 100% open source, easy to customize and secondary development
- Rich API interface for easy integration into existing systems
In addition, Bright Data's MCP servers are equipped with powerful data processing capabilities that not only efficiently capture web content, but also analyze and clean the data in real time to ensure that the information obtained is accurate.
This technology is widely used in market research, public opinion monitoring, e-commerce and other fields to provide accurate data support for enterprises.
For more information, please visit: https://t.co/n2WWvlbxEm
---
Meta has just released the new KernelLLM 8B model on the Hugging Face platform.
The model outperforms well-known models such as GPT-4o and DeepSeek V3 in single-trial inference in the KernelBench-Triton Level 1 benchmark. Specifically, the performance of KernelLLM 8B in a single inference task is impressive.In multiple reasoning tasksThe KernelLLM 8B performs just as well and even outperforms the DeepSeek R1.
This breakthrough not only demonstrates Meta's technological strength in the field of large language modeling, but also provides the industry with a new high-performance option.KernelLLM 8B has a wide range of application scenarios, including natural language processing, machine translation, text generation, and other fields.
Meta said they will continue to optimize the KernelLLM 8B model and plan to introduce more innovative features in the future to meet the growing demand for artificial intelligence.
---
Mistral Medium 3 scored significantly in the community in the latest assessment.#11 The overall ranking of the Mistral Medium 3 improved it by more than 90 points over the Mistral Large. In addition, the Mistral Medium 3 performed particularly well in technical areas such as math, hard cues, and coding, placing 5th and 7th, respectively. It also finished 9th in the web development arena.
Mistral Medium 3 is a new multimodal model that offers state-of-the-art performance and an 8x cost reduction compared to other models. This model excels in balancing performance, cost, and deployment flexibility, and is especially highly capable in coding and function calls. In addition, it comes with comprehensive enterprise-grade features, including hybrid or local/VPC deployment, customized post-training, and seamless integration with enterprise tools and systems.
The MistralAI team congratulates this achievement and looks forward to more surprises in the next version of Mistral Large.
For more details, visit the official MistralAI blog.
---
Recently.PixVerse V4.5version has achieved impressive results in the iOS App Store. After only one week online, it ranked fourth on the overall charts and first in the photo video category.
The update brings more than 20 cinematic lens motion effects, multi-subject referencing and blending capabilities, and support for complex actions. These new features not only enhance the user experience, but also provide content creators with more creative possibilities.
- Cinematic Lens Motion: A variety of new cinematic lens motion effects have been added to give ordinary videos the texture of a professional movie.
- Multi-subject reference and fusion: Users can fuse the movements and gestures of multiple subjects to create a more natural and smooth effect.
- Complex Motion Support: Improved handling of complex motions, allowing more creativity to be realized.
In addition, some users shared 8 wonderful examples they created using PixVerse V4.5, demonstrating the application's powerful features and wide range of application scenarios.
As mobile video editing tools continue to evolve, PixVerse V4.5's strong performance proves its leadership in the market.
---
Alibaba's Wan team recently released Wan 2.1-VACE, a unified AI model designed for video creation and editing. The model is available in two different scales, 1.3B and 14B, and supports a variety of features, including reference image to video, video to video, and video editing with masks.
Wan2.1-VACE technically enables a seamless transition from single input to complex video generation and is capable of handling a wide range of video editing tasks for content creators, filmmakers, and video editors in professional fields. In addition, the model has been open-sourced under the Apache 2.0 license, allowing developers to use it freely and develop it further.
Key Features:
- Support reference image to video generation
- Video to video editing support
- Support for video editing with masks
- Available in 1.3B and 14B size versions
The launch of Wan2.1-VACE will further promote the automation and intelligence process in the field of video creation and editing, bringing new changes to the industry.
---
Starting today, you can access over 5,000 MLX-powered Large Language Models (LLMs) directly on the Hugging Face Hub. Simply click the "Use this model" button on any compatible model page to quickly launch an OpenAI-compatible server.
This new feature greatly simplifies the process for developers, eliminating the need for complex configuration and deployment steps to get efficient and intelligent services on the endpoint. In addition.Users can also launch OpenAI-compatible servers directly from the model page, further enhancing the ease of use.
Whether performing tasks such as natural language processing, machine translation or text generation, this new feature provides developers with powerful tools to support them.
- Visit Hugging Face Hub
- Select Compatible Models
- Click on "Use this model"
- Starting an OpenAI-compatible server
In the future, we look forward to seeing more innovative applications and features rolled out.
