SSkywork Super Agents

Skywork says its deep research framework scored 82.42 on GAIA, a benchmark for general AI assistants, and ranked first on the Hugging Face leaderboard as of May 10, 2025, ahead of OpenAI Deep Research and Manus. It launched Skywork Super Agents worldwide on May 22, 2025, claiming research quality on par with OpenAI’s at about 40% of the cost. The pitch is an AI office suite: one prompt becomes a sourced report, deck, spreadsheet, webpage or podcast.

Skywork Super Agents suits analysts, consultants, students and small teams who need research turned into finished files and want every claim traceable. Cost visibility is the main thing to plan for: third-party reports put Pro at about $19.99 a month, and Skywork could help buyers by publishing how many credits each task uses. We rate it 8.2/10 on our scale. We have not used Skywork ourselves; this review is based on Skywork’s launch announcement, third-party pricing analysis and public reporting on its parent company.

Benchmark Claims and the Kunlun Tech Backing

Skywork launched with benchmark claims and has since become part of a large AI business at Kunlun Tech, which gives buyers some signal on staying power.

The benchmark numbers are the headline, but they are company reported and from May 2025, and agent leaderboards change quickly. Treat them as evidence that the research engine was competitive at launch, not as a current ranking.

The parent company figures matter more for a buyer choosing a long-term tool. A tools business above $100 million in ARR, inside a group reporting $800 million across AI products, suggests Skywork is a funded product line rather than a side project.

Agents Sharing One Deep Research Engine

Skywork Super Agents is a web workspace with several agents that share one research engine, built on Skywork’s own deep research model.

Scoping the Task

You type a request and pick an agent, or use the general agent. Before starting, Skywork shows a Clarification Card: fill-in-the-blank and multiple-choice questions about scope, audience and format. Your answers become a to-do list that you confirm, which reduces wasted runs on a misunderstood brief.

Research and Sources

The agent then researches the web and any personal knowledge bases you attach. A knowledge base can hold up to 50 documents, including PDFs, Word and PowerPoint files, spreadsheets, audio and links, and you can keep several. Every document, deck and sheet keeps its references, so readers can trace a claim back to its source.

The Specialist Agents

• Documents: research reports with bar charts, histograms, line graphs, pie charts, radar plots and data tables.
• Slides: decks built from the research, editable online and exportable to PowerPoint and PDF.
• Sheets: descriptive and inferential statistics on uploaded data, charts and summary tables, exportable to Excel.
• Webpages: interactive sites such as blogs or simple product pages, without code.
• Podcasts: a script from a short prompt with synthesized audio, in English at launch.

A general agent uses dozens of MCP tools for images, posters, music, music videos, promotional videos, audiobooks and illustrated books. Third-party reporting from July 2026 also describes an in-product agent called SkyClaw that builds slides, documents, sheets, sites and media and can run scheduled tasks.

A Typical Run

Take a market overview for a client meeting. You ask the Documents agent for a report on a sector, answer the Clarification Card questions about audience, length and regions, and approve the to-do list. The agent researches, writes the report with charts and tables, and attaches its references. From there you can ask the Slides agent to turn the same research into a deck, or upload a spreadsheet to the Sheets agent for summary statistics and charts. Because the agents share the research engine and your knowledge bases, the outputs stay consistent with each other, and you can review each file before it leaves the workspace.

For Developers

Skywork published its deep research agent framework as open source on GitHub and Hugging Face, and offers an MCP server that lets developers generate documents, slides and sheets from their own tools.

Output Files, Not Actions in Your Business Systems

The agents produce files; they do not log into your CRM, send emails or update tickets. A person still checks sources, corrects numbers and decides what goes to a client. Benchmarks measure research accuracy on test questions, not the accuracy of your specific report.

Credit-Based Plans From Third-Party Analysis

The figures below come from third-party analysis published in July 2026; check them at signup. All plans run on credits, and heavier tasks such as deep research or long video renders use more credits than quick edits.

Pro Monthly Versus Annual for a Consultant

A consultant using Pro all year: monthly billing at $19.99 comes to $239.88 a year. The reported annual plan at $149.99 saves $89.89, or works out to about $12.50 a month.

A four-person research team: four Pro subscriptions at $19.99 a month cost $79.96 a month, or $959.52 a year. On annual plans, the same team would pay $599.96 a year.

Light users: someone producing a few reports a month may stay on the free tier, though free credits are limited and do not roll over, according to third-party analysis.

The variable cost is credits. Because Skywork does not publish a credit cost per task, the first month on any plan is effectively a measurement exercise. Run a few typical jobs, check how many credits each uses, and only then commit to an annual plan. Watch the $1 trial, which is reported to convert to a paid plan on day seven.

Skywork Super Agents Compared With Manus, Genspark Super Agent, ChatGPT Work and Gamma Agent

General agents now overlap heavily. The differences come down to research depth, output formats and how costs are metered.

Skywork’s edge is source tracing and research depth for document-style work. Manus and Genspark cover similar ground at similar prices; ChatGPT Work wins on convenience for existing ChatGPT users; Gamma is better if design matters more than research. See more in our general-purpose AI agents and AI research agents pages.

Analysts, Consultants and Researchers

• Analysts and consultants: reports and decks arrive with traceable sources, which shortens fact checking before client delivery.
• Students and researchers: knowledge bases let you ground the agent in your own papers and notes, up to 50 documents per base.
• Small teams producing data summaries: the Sheets agent runs statistics and charts on uploaded data without a separate analysis tool.
• Marketers and content teams: one brief can produce a report, a deck and a simple landing page that share the same research, which keeps messaging consistent across formats.
• Developers: the open-source framework and MCP server let you build document, slide and sheet generation into other tools.

Teams Needing CRM Actions or a Fixed AI Budget

• Teams that need actions in business systems: look at agent platforms that connect to CRMs and help desks instead.
• Buyers who need a fixed AI budget: per-task credit costs are not published, and long jobs consume more.
• Design-led presentation work: Gamma Agent or Canva AI give more layout control.
• Non-English podcast production: the Podcasts agent launched in English only.

Pricing Details, Benchmark Dates and Source Checks

Pricing transparency is the main area to improve. Most plan details come from third-party analysis, and the same analysis describes Skywork’s own pricing guide as partly out of date. Check credits, allowances and renewal terms on the checkout page before paying, and set a reminder for the $1 trial conversion.

Benchmarks age fast. The GAIA ranking is from May 2025 and was reported by Skywork itself. Newer agents from OpenAI, Google and others have shipped since, so judge Skywork on a trial with your own tasks.

Source tracing helps, but it does not guarantee accuracy. A cited source can be weak or misread. Open the references behind any number that will reach a client or a decision.

The output formats have a defined scope too. Webpages builds simple sites such as blogs and product pages, not full web applications, and podcast audio launched in English only. Teams that need polished, brand-controlled design will likely still finish decks in another tool.

Data handling deserves attention. Knowledge bases can hold sensitive documents, so confirm how uploads are stored and whether they are used for training before adding client material.

The May 2025 Global Launch and Kunlun’s 2026 Update

On May 22, 2025, Skywork launched Super Agents worldwide with five specialist agents, an open-source deep research framework and an MCP server for developers.

On July 2, 2026, Kunlun Tech reported more than $800 million in ARR across its AI products for the second quarter, including over $100 million from its AI tools business, which includes Skywork.

On July 20, 2026, an eesel analysis of Skywork pricing reported a $1 trial, Pro at about $19.99 a month and an annual plan at about $149.99.

Strong for Work That Ends in a Doc, Deck or Sheet

Skywork Super Agents is a strong pick for anyone whose work starts with research and ends in a document, deck or spreadsheet. The Clarification Card cuts wasted runs, sources stay attached to the output, and the reported Pro price is in line with Manus and Genspark. A free tier makes it easy to test, and the open-source research framework gives developers a way to inspect and reuse the engine behind it, which few rivals offer.

The 8.2/10 on our scale reflects areas Skywork could build on: published credit costs, first-party pricing details, actions in business systems and more public reviews. Try it on two or three real tasks, measure the credits each uses, and compare the output side by side with Manus or Genspark before choosing a paid plan.

Loose Ends on Skywork Credits and File Output

How Much Does Skywork Super Agents Cost?

There is a free tier with limited credits. Third-party analysis reports a $1 seven-day trial, a Pro plan at about $19.99 a month and an annual plan at about $149.99. Credit costs per task are not published.

Did Skywork Really Top the GAIA Benchmark?

Skywork reported that its deep research framework scored 82.42 on GAIA and ranked first on the Hugging Face leaderboard as of May 10, 2025, ahead of OpenAI Deep Research and Manus. The ranking is company reported and from 2025, and leaderboards change often.

Can Skywork Export Slides to PowerPoint?

Yes. The Slides agent produces decks you can edit online and export to PowerPoint or PDF. The Sheets agent exports to Excel.

Does Skywork Super Agents Have a Free Plan?

Yes. A free tier lets you try the agents with a limited credit allowance that, according to third-party analysis, does not roll over. Paid plans add a larger allowance, and a $1 seven-day trial is reported to convert to a paid plan unless you cancel.

How Is Skywork Different From Manus?

Both handle many task types from one prompt at similar prices. Skywork leans on research depth and keeps sources attached to documents, slides and sheets, while Manus covers a wider range of tasks, including simple web apps.