Skywork Deep Research Agent Major Upgrade: Delivering Enhanced Multimodality, Superior Output Quality, and Optimized Efficiency

SINGAPORE, Aug. 14, 2025 /PRNewswire/ -- The SkyWork AI Technology Release Week officially kicked off on August 11. From August 11 to August 15, SkyWork will release one new model each day for five consecutive days, covering cutting-edge models for core multimodal AI scenarios. As of now, we have already launched the SkyReels-A3, Matrix-Game 2.0, Matrix-3D, and Skywork UniPic 2.0 models.

On August 14, Skywork Deep Research Agent v2 was officially launched as the core engine of Skywork Super Agents.

Since its initial launch on May 22, Skywork Deep Research Agent has significantly reshaped the role of large language models in the AI Office space. Through the skywork.ai platform, it has produced a vast number of high-quality documents, PowerPoint presentations, spreadsheets, and other deliverables with exceptionally high information density for users. The newly upgraded Skywork Deep Research Agent v2 introduces the following enhancements to the user experience.

Users worldwide are welcome to register and use skywork.ai:
Global website: https://skywork.ai 

1 "Multi-Modal Deep Research" Agent – The First to Integrate Multi-Modal Retrieval, Understanding, and Generation

Existing Deep Research Agents in the industry rely exclusively on searching and scraping textual data from web pages, restricting their analysis to plain text. However, more than half of the internet's critical information exists in mixed text-and-image formats, such as financial report graphs, experimental diagrams in research papers, social media comparison visuals, and proposal flowcharts.

Overlooking such multi-modal data deprives the Agent of key decision-making insights, significantly compromising output quality. To address this challenge, the Skywork team has launched the industry's first Multi-Modal Deep Research Agent by seamlessly combining multi-modal retrieval, understanding, and cross-modal generation into deep research workflows.

This feature is now live on skywork.ai (https://skywork.ai) and available to users worldwide.

To enhance multi-modal information retrieval capabilities, the Skywork team has pioneered four technological breakthroughs, including MM-Crawler (multi-modal crawler) technology, long-context multi-modal information aggregation, asynchronous parallel processing with multi-agent understanding architecture, and multi-modal output generation.

Through these technological innovations, the multi-modal Skywork Deep Research Agent v2 has finally accomplished what seems simple yet was long neglected—simultaneous text reading and image comprehension. Therefore, it enables researchers and users to generate comprehensive, logically structured, and visually refined in-depth reports in a single step.

2 "Multi-Modal Deep Browser Agent" – Redefining Social Media Analytics and Data Intelligence

To deliver capabilities unmatched by conventional browsers, including ultra-low latency, guaranteed high response rates, optimized task completion, and adaptive decision-making, the team has implemented critical proprietary advancements in the Skywork Browser Agent, covering enhanced DOM + visual reasoning architecture, native integration with major platforms, Parallel Search technology, Multi-Action planning mechanism, intelligent filtering, seamless human-AI collaboration, privacy protection and security compliance.

The Skywork Browser Agent now achieves human-like browsing and interaction capabilities, fundamentally transforming traditional approaches to data collection and analysis. The agent demonstrates remarkable precision and efficiency in executing intelligent search operations, performing multimodal information analysis, and deriving actionable insights from community content. By effectively resolving limitations inherent in conventional browser agents, it showcases the significant potential of Skywork Super Agents in handling both long-horizon tasks and vision-language actions (VLA).

The Skywork Browser Agent has entered its alpha and invite-only testing phase, with full public release expected soon for all skywork.ai users.

Skywork Browser Agent's core capabilities:

1. Advanced multimodal comprehension: Going beyond text-only analysis, it achieves deep semantic understanding of social media content—including platforms like Xiaohongshu, Twitter, and Instagram—by extracting insights from images/videos and analyzing comment sentiment, enabling holistic data intelligence.
2. Automated data analysis & reporting: The agent performs efficient community content analysis and transforms raw research data into intuitive, visually digestible reports.
3. One-click website generation: The agent automatically curates key visuals, analyzes their content, and deploys them as ready-to-use standalone websites with one click, thereby streamlining result presentation and team collaboration.
4. Seamless workflow integration: It is designed for interoperability with information retrieval agents and document tools (e.g., PPT/Doc assistants). When drafting reports, it intelligently retrieves and recommends relevant visual assets, thus dramatically boosting productivity.

3 Achieving SOTA Across Various Benchmarks with Enhanced Deep Information Retrieval & Complex Task Execution Capabilities

To enhance the foundational model's performance in complex task execution, including advanced information retrieval, synthesis, and summarization, Skywork Deep Research Agent v2 integrates multiple breakthrough mechanisms: high-quality synthetic data generation and curated training, end-to-end reinforcement learning, highly efficient parallel inference, and multi-agent self-evolution frameworks. Benchmark evaluations confirm its superior performance, setting new state-of-the-art (SOTA) results industry-wide.

On the authoritative search evaluation benchmark BrowseComp, Skywork Deep Research demonstrates exceptional performance. In standard mode, it already surpasses most competing solutions with an accuracy rate of 27.8%. When activating its proprietary Parallel Thinking mode, the accuracy jumps significantly to 38.7%, setting a new industry SOTA record.

Notably, in Parallel Thinking mode, Skywork Deep Research's accuracy exhibits continuous improvement with extended processing time, which demonstrates the exceptional scalability and untapped potential of our proprietary system architecture.

The API preview feature is now available. To request access, please visit Skywork's official GitHub repository and submit your application:
https://github.com/SkyworkAI/Skywork-DeepResearch

In addition, Skywork Deep Research Agent has achieved SOTA performance on the GAIA Test benchmark, which validates its advanced capabilities in complex task execution

Skywork Deep Research Agent v2 will soon launch comprehensively across all deep research applications on skywork.ai.

SOURCE Skywork AI pte ltd