Skywork AI

Skywork AI is built by a Singapore-based company backed by Kunlun Tech, a publicly traded Chinese tech conglomerate (SZSE: 300418) that invested 500 million yuan (roughly $70 million) in Skywork as an A-round in August 2025.

The platform launched globally on May 22, 2025, and claims over 100 million users, though I could not independently verify that number.

The site receives significant traffic, and the product has been compared directly to Manus and Genspark as part of the emerging “AI agent workspace” category.

The core idea is simple: you describe what you need, Skywork researches it, and then its specialized agents produce the deliverables.

The platform calls this the “Super Agent” architecture. There is a General agent for open-ended prompts, and then dedicated agents for Documents, Slides, Sheets, Websites, Posters, Designs, Videos, and Podcasts. A “Tools Agent” handles miscellaneous tasks.

I tested Skywork by giving it a research-heavy project: a competitive analysis of the AI writing assistant market, including a report, a 10-slide summary deck, and a comparison spreadsheet.

The platform started with a “Clarification Card” – a set of follow-up questions asking me to specify the market scope, time frame, and which competitors to include.

This is a good design choice. Most AI tools just guess and start generating. Skywork asked three targeted questions and then confirmed a task list before running.

The deep research phase took about four minutes.

The resulting report was 2,800 words with 23 inline citations, and the citations were real – I spot-checked eight of them, and seven linked to actual pages with the referenced data. One link was dead.

That is a better citation hit rate than I have seen from Perplexity’s deep research or ChatGPT’s web browsing on a task of similar scope.

The slide deck generated from the same research was mixed.

The structure was logical – market overview, competitor profiles, pricing comparison, trend analysis – and the data was consistent with the report.

But the design was generic.

Every slide had the same layout.

The charts were readable but not something I would put in front of a client without reformatting in PowerPoint.

Skywork exports to PPTX and PDF, so the editing path exists, but “AI-generated slides that need a design pass” is not a unique value proposition in 2026.

The spreadsheet was the strongest secondary output. It pulled the pricing data from the research into a structured comparison table with formulas for annual cost calculations.

The formatting was clean enough to share directly.

Key Features

DeepResearch Engine

This is what separates Skywork from tools like ChatGPT or Claude that do web searches as a side feature.

Skywork’s research agent reportedly uses reinforcement learning to decide which sources to pursue, scans hundreds of pages per task, and cross-references findings before synthesizing.

The company claims 98.7% citation accuracy and a #1 ranking on the GAIA benchmark (82.42 as of May 2025).

These are vendor claims, not independently audited, but in my testing the citation quality was noticeably higher than what I get from a standard ChatGPT web search.

The research output read like a junior analyst’s first draft rather than a chatbot’s summary – it had structure, sourcing, and a narrative thread connecting the findings.

Documents Agent

A two-panel writer that produces long-form content with tone adjustment and version history.

The output includes inline citations from the research phase.

I found the document quality solid for first drafts of reports and briefs, though the writing style leans corporate-neutral and needs editing for any audience-specific voice.

Slides Agent

Generates decks from research or uploaded content, with page-by-page editing.

Exports to PPTX and PDF. The research-backed charts are a strong point – the data is pulled from cited sources, which means you can trace a chart back to its origin.

The design, as noted, is generic.

If your workflow already includes a design pass in PowerPoint or Canva, this saves the content structuring step. If you need presentation-ready slides from an AI tool, Gamma does this better.

Sheets Agent

Chat-driven spreadsheet generation and analysis.

You describe the data you want, and Skywork builds it from its research or your uploads. Exports to XLS and Google Sheets.

The formulas were correct in my testing, and the data organization was logical. This was the output that needed the least editing.

Website Generator

Generates React code and publishes to a Skywork subdomain or custom domain, with built-in analytics and SEO tools.

I did not test this feature in depth, but it exists for users who want to turn a research project into a landing page without switching to a separate site builder.

Video and Image Generation

Skywork includes an image canvas (using models like GPT Image 2 and Nano Banana Pro) and a video agent with a multi-track timeline. The video agent supports models including Sora, Kling, and Wan.

Video output ranges from 5 to 10 seconds at up to 1080p. This is an add-on capability, not a core strength – dedicated video tools like Runway or Kling produce higher-quality output.

Knowledge Bases

You can upload up to 50 documents (PDF, DOC, PPT, XLS, audio, or URLs) per knowledge base, and the agents use them as context for generation. This is useful for grounding output in your own data rather than relying entirely on web research.

Automation and Integrations

Scheduled tasks, MCP connections, and bots for Telegram, WhatsApp, Discord, Slack, Line, and Feishu/Lark.

The 24/7 persistent host on paid tiers lets you set up recurring research or content generation tasks. I did not test the bot integrations.

Competitors Comparison