Genspark Super Agent 2

• 1. overview (612 chars)

  Genspark Super Agent is a "one-stop AI workspace" created by MainFunc (external brand Genspark / GENSPARK INC). The core is a general-purpose agent that can understand natural language intentions, disassemble tasks by itself, and deliver finished products end-to-end. It was originally an AI search engine in 2024, and fully shifted to the agent route in April 2025. By 2026, it has been iterated to AI Workspace 4.0, and extended "AI employees" to the local desktop and Microsoft Office. What Daily Product Express calls "Super Agent 2" corresponds to its second-generation App (2.0.x series) and multiple rounds of Workspace version evolution from 2025 to 2026; it is essentially Genspark's flagship Super Agent platform.

• 2. Product background (1241 chars)

  Genspark was founded by Eric Jing, the former vice president of Baidu and the former CEO of Xiaodu Technology. He was the head of Microsoft's Bing Asia market research and development and the inventor of Microsoft Xiaoice. Co-founder and CTO Zhu Kaihua was the chief architect of Google, chief architect of Baidu Intelligent Assistant and CTO of Xiaodu. The team has both "China engineering density" and "Silicon Valley product rhythm", and there were only about 20 people in the early days.   The company’s financing is at a very fast pace: US$60 million in June 2024; US$100 million in Series A in February 2025 (post-money valuation of US$530 million); US$275 million in Series B from November 19 to 20, 2025 (lead investor Emergence Capital, followed by SBI and LG Technology Ventures, with a valuation of US$1.25 billion), and simultaneously announced access to Microsoft Agent 365 Ecology; received another US$385 million expansion round in April 2026, with the valuation rising to US$1.6 billion. As of April 2026, the official cumulative financing has reached approximately US$545 million (including Temasek Pavilion Capital, etc.).   The growth is equally fierce: the annualized revenue (ARR) in October 2025 will be approximately US$155 million, and will exceed US$250 million from March to April 2026, with more than 2 million monthly active users and approximately 100,000 paid seats. It has been compared by many media to be close to the early growth curve of OpenAI.

• 3. Product functions and usage experience (1969 chars)

  The skeleton of Super Agent is the "Mixture-of-Agents (MoA)": a central orchestrator dynamically schedules 70+ top models according to task complexity (Claude is responsible for planning and agent reasoning, GPT-type models are responsible for in-depth research and writing, Gemini is responsible for multi-modal understanding, DeepSeek and other open source models are in charge), and calls 80+ built-in tools and 10+ own data sets. Multiple models verify each other, with the goal of reducing illusions.   The implementation capabilities cover a wide range: automated multi-step in-depth research, "Sparkpages" interactive knowledge page with references (embedded video clips, comparison tables, pictures, you can continue to ask), generate a set of editable PPT in about 4 minutes, convert about 5 hours of YouTube videos into 10 pages of PPT with one click, website building, table data analysis (real-time capture and automatic writing of Python visualization), real phone dialing (Call for Me, third-party test completion rate of 47 times) 83%), and image/video/audio generation. Its "backtracking" capability allows it to self-recover when substeps fail, rather than stopping when an error occurs.   The Workspace iteration in 2026 will spread the "AI employees" more widely: Workspace 2.0 adds voice input Speakly, AI music and AI audio agents; 4.0 (April 8, 2026) is a platform-level jump-Genspark Claw for Desktop native client can directly operate your local files and browser (Computer Use + Browser Use, no need to upload) for the first time, and Microsoft Office native plug-in Slides/Sheets/Docs agents are embedded in PowerPoint, Excel, and Word; Advanced Workflows uses the OpenCode engine to run more complex multi-step automation; Speakly adds real-time translation and Meeting Bots that automatically join meetings and take notes; Ultra mode is also connected to Claude Opus 4.7.Objectively speaking, its experience lies in the "finished product in one sentence": Compared with opening twenty browser tabs to conduct competitive product research, Sparkpages' structured documents with video timestamps and comparison tables really save time. Weaknesses are that the depth is not uniform enough, complex tasks consume a lot of points, and some advanced abilities still occasionally roll over.

• 4. Pricing and business model (891 chars)

  Genspark adopts a points subscription system and stacks team seat pricing. There are several public versions, based on the mainstream evaluation in 2026: the free version is about 100-200 points per day, with unlimited basic searches; the Plus version is about $19.99/month (annual payment), about 10,000 points/month; the Pro version is about $199/month (annual payment), about 125,000 points/month; the team/enterprise version is about $30/seat/month. AI chat and standard image generation will not consume points until the end of 2026, while videos, complex agent operations and dialing will only consume points. Points will be cleared at the end of the month and will not be rolled over.   The business model is B2B SaaS first: first use free files to achieve viral growth (users share the finished product "Remix with Genspark" with colleagues to bring in new customers), and then convert payment to teams that require advanced capabilities and collaboration. MoA also drives down the cost structure by routing tasks to the most cost-effective model.

• 5. User reviews (893 chars)

  The positive feedback focused on "can really get the job done for me." A management consultant in Boston said in September 2025 that the 20-page market analysis report generated by it would have required a junior analyst to do it for a whole week; Newton Kalani on Product Hunt commented that "the results are highly relevant and accurate, and AI seems to understand the context of search terms particularly well." Calling on behalf of others, PPT and Sparkpages are frequently mentioned useful scenarios.   Negative feedback is highly concentrated: Trustpilot score is only about 1.9/5 at the beginning of 2026, and negative reviews mainly point to three points - opaque billing (accused of switching monthly payments to annual payments without consent, charges still being deducted after logging out, and no self-service cancellation button in the settings), confusing points consumption (a single task consumes 10-500 points, no estimate in advance, and no rollover), and some functions have limited depth and cannot stably replace key business processes.

• 6. Industry evaluation (749 chars)

  When the GAIA benchmark (which measures the ability of an agent to solve complex real-life problems) was released, Genspark Super Agent Level 1 scored 87.8%, Level 2 scored 72.7%, and Level 3 scored 58.8%. All three scores were higher than Manus and OpenAI Deep Research, making it the SOTA at the time. The industry generally regards it and Manus as the most direct rivals in the "autonomous agent" track: Genspark's differentiation lies in the breadth of tools, Microsoft binding and a more aggressive release rhythm (Workspace 3.0 to 4.0 are only three weeks apart).   The threats are equally clear: Microsoft 365 Copilot ($30/seat) and Google Gemini Workspace ($14/seat) can build a moat by leveraging existing corporate relationships; Notion, Coda, and Canva are also adding autonomous agents; ChatGPT and Perplexity are contrasting on the conversation and search sides respectively.

• 7. Controversies and Risks (687 chars)

  The most prominent controversy is the billing and canceling experience. Many reviews and communities have listed it as "not suitable for critical business flows". Secondly, there is the "hidden threshold" of the points economy: it seems that there is a free file, but complex tasks quickly burn up the points and do not clear them, and the trial transfer fee is not user-friendly. Technically, despite MoA mutual calibration and backtracking, long tasks still have the probability of hallucinations and execution failures, and physical world actions such as real phone calls only have about an 80% completion rate. At the data level, the App's privacy statement shows that it may collect search history, device ID, etc. for analysis and function improvement. Cross-border users need to pay attention to compliance.

• 8. Applicable groups and usage suggestions (698 chars)

  It is suitable for people who are overwhelmed by repetitive and multi-step office tasks - making PPT, research and abstracts, publishing the first version of graphics and videos, making appointments, and organizing meeting minutes. The free version has 100-200 points per day, which is low-risk enough for a trial. It is recommended to do a stress test with a real task (such as making a report PPT or making a restaurant reservation) first, and let the results judge whether it is worth it for you.   If you mainly want conversation, writing and instant Q&A, ordinary chatbots are cheaper and simpler; if you are deeply integrated into the Microsoft/Google ecosystem, Copilot and Gemini have lower migration costs. Think of Genspark as an "on-demand execution layer" instead of a "daily chat layer", which is the most cost-effective.

• 9. Summarize (396 chars)

  Genspark Super Agent packages "hybrid agents + more than 80 tools + real-world actions" into an AI employee that can deliver finished products. In the autonomous agent track, the breadth of tools and growth rate are both in the first echelon; the real lessons that need to be made up are bill transparency, predictable points and the stability of key processes. Otherwise, no matter how fast the growth is, it will not escape the ceiling of "strong experience and weak trust".

No in-depth report yet

• 

  silverelephant680

  —

  回不去了，真回不去了。

---

• 

  CCookX

  —

  写作风格有点重复，prompt 不写细就一股模板味。

---

• 

  OliviaGarcia_99

  —

  卡成 PPT，高峰期偶尔转圈。

---

• 

  小鱼427

  —

  免费档每天 100 多积分够我试两周了，先白嫖再说。

---

• 

  邹莉心

  —

  Credits 烧得有点凶，复杂任务一趟就扣好几百，事前还没预估，新手很容易还没搞明白就见底了。建议先用免费档摸清楚再上付费。

---

• 

  薛刚海

  —

  视频和图片生成还行，但深度比不过 Runway 和专门的工具。把它当研究加出初稿的瑞士军刀，别当专业创作软件，心态就对了。

---

• 

  trueMaanasNamnaik_pro

  —

  好处是真能替你把活干完，坏处是深度不均。简单调研封神，遇到特别专业的统计类题目引用就开始发飘，得自己再核一遍。

---

• 

  Maria.Edwards_2022

  —

  Trustpilot 才一点几星，评论几乎一边倒吐槽扣费，吓退我了。

---

• 

  ElenaRajković

  —

  中文支持还成，但长文档偶尔掉链子，复杂表格不如自己手搓。

---

• 

  KathleenRoberts_Plus1

  —

  Genspark 的幻灯片真的是一绝，单这条就值回票价了。

---

• 

  TokelTiger499

  —

  说句实话，Genspark 最让我惊喜的是 MoA 多模型路由。同样一个问题，它把 GPT、Claude、Gemini 各跑一遍再互相校验，出来的东西比单独用哪个模型都稳。19.99 美元年付那档对重度用户是真划算，等于把 ChatGPT、Claude 和图像工具一股脑塞进一个面板了。

---

• 

  JUwar

  —

  Claw 桌面端出来之后顺手多了，直接读我本机文件不用上传，还能自己点浏览器填表抓数据。配合 Office 插件在 PPT 里直接出稿，基本把我从复制粘贴里解放出来了。唯一担心是它看得到我屏幕和文件，隐私这块得自己权衡。

---

• 

  Jessica_AllenIII

  —

  我们小团队把 ChatGPT Team、Midjourney 和 Jasper 都退了换 Genspark Pro，一年省了四千多刀授权费。研究、出图、做 deck 一把梭，唯一别扭的是导出偶尔抽风，PDF 和 PPT 有几次没顺利下来，得重试。

---

• 

  Fla_mingoFiKrauss

  —

  我觉得它值不值完全看你怎么用。轻量问答和聊天老实用 ChatGPT 更便宜更稳；但如果你经常要从一个问题跑到一份带引用的报告加一套 PPT，它就是能把几小时压成几分钟。关键别拿它当默认工具啥都丢过去，积分会教你做人。

---

• 

  DBaker_599

  —

  账单这块是真的劝退。我朋友按月付试用，结果被悄咪咪切成年付，注销账号后还扣了一笔，设置里翻半天也找不到自助取消的按钮。这种不透明对想尝鲜就走的人非常不友好，建议用虚拟卡或者试完立刻留意扣费。

---

• 

  ClementeGuajardo

  —

  我用它做竞品调研，以前要开二十个标签页翻一下午，现在一个 Sparkpage 出来带引用带对比表，省了起码三小时。给分析师当第一手草稿很香。

---

• 

  99evwm8m

  —

  GAIA 那个 87.8% 分数我是服的，实际用下来多步任务确实不容易卡死，子步骤失败了还会自己绕着走，比那种直线流水线稳。

---

• 

  DennisCook

  —

  电话按秒烧积分，三分钟通话就是 180 积分，Plus 那一万额度折算下来也就五十来通。偶尔预约还行，真拿它当外呼团队用不起。

---

• 

  Linda_Hall_X

  —

  代拨电话那一下真的有点科幻，让它帮我订餐厅居然真打通了。

No reviews yet