Draft:Agent S
 | Review waiting, please be patient.
This may take 8 weeks or more, since drafts are reviewed in no specific order. There are 3,868 pending submissions waiting for review.
Where to get help
How to improve a draft
You can also browse Wikipedia:Featured articles and Wikipedia:Good articles to find examples of Wikipedia's best writing on topics similar to your proposed article. Improving your odds of a speedy review To improve your odds of a faster review, tag your draft with relevant WikiProject tags using the button below. This will let reviewers know a new draft has been submitted in their area of interest. For instance, if you wrote about a female astronomer, you would want to add the Biography, Astronomy, and Women scientists tags. Editor resources
Reviewer tools
|
Comment: In preparing this draft, I disclose that AI assistance was used as follows: The initial draft was written with Sai, an AI agent made by Simular (my employer, which also develops Agent S). I chose the sources from an earlier source review, and Sai wrote the wikitext from them. I read it through and asked Sai for changes, including plainer, less promotional wording and replacing self-published sources with conference listings where possible. I then checked the claims against the cited sources. The chat transcript is not publicly shareable. Samgyw (talk) 09:11, 2 November 2026 (UTC)
 | This article has multiple issues. Please help improve it or discuss these issues on the talk page. (Learn how and when to remove these messages)
|
| Agent S | |
|---|---|
| Developer | Simular |
| Release | October 2024 |
| Written in | Python |
| Type | AI agent framework |
| License | Apache License 2.0 |
| Repository | github |
Agent S is an open-source framework for building computer-use agents. These are programs that operate a computer the way a person does, by reading the screen and using the mouse and keyboard.[1][2] It is developed by researchers at the startup Simular.[3] Four versions have been released since October 2024. Each is described in a research paper and tested on the OSWorld benchmark.[4][5] The code is published on GitHub under the Apache License 2.0.[6]
History
[edit]Agent S
[edit]Saaket Agashe, Jiuzhou Han, Shuyu Gan, Jiachen Yang, Ang Li and Xin Eric Wang posted the first Agent S paper to arXiv on 10 October 2024.[1] It was accepted as a poster at the International Conference on Learning Representations (ICLR) in 2025.[7] The Decoder covered the release that month and noted that the Python code was free to download from GitHub.[2]
Agent S2
[edit]The Agent S2 paper followed on 1 April 2025.[8] Wired covered it that month, focusing on how the agent switches between AI models depending on the task.[3] The paper was accepted at the Conference on Language Modeling (COLM) in 2025.[9] In July 2025, The Register reported that Agent S2 ranked fourth on the OSWorld leaderboard and sixth on AndroidWorld.[10]
Agent S3
[edit]Agent S3 came out in October 2025, in a paper titled "Scaling Agents for Computer Use".[11] The Chinese technology publication 机器之心 (Synced) reported that it was built directly on Agent S2, with a simpler framework and a built-in coding agent.[4] The paper was accepted by the journal Transactions on Machine Learning Research (TMLR) in 2026.[12]
Agent S4
[edit]The Agent S4 paper, "Evolving Agent Interfaces", was posted on 20 October 2026.[5]
Architecture
[edit]The first version breaks a task into smaller steps before acting. To plan those steps, it searches for outside knowledge and looks up how it handled similar tasks before. Its authors call this experience-augmented hierarchical planning.[1][2] It also uses an agent-computer interface (ACI), which the authors designed to help agents built on multimodal large language models reason about the screen and control it.[1] An independent 2025 survey of GUI agents describes Agent S as turning user instructions into detailed subtasks, ordered so that each step comes after the steps it depends on.[13]
Agent S2 splits the work between a general model and specialist models. The paper uses a technique called Mixture-of-Grounding to find elements on the screen precisely. A second technique, Proactive Hierarchical Planning, revises the plan at several time scales as the screen changes.[8] Simular co-founder Ang Li told The Register that once the agent finds a way to do a task, it turns those actions into symbolic code that runs the same way each time. When that code stops working, the agent goes back to the language model.[10]
A computer-use agent can succeed at a task on one run and fail on the next. Agent S3 runs several attempts in parallel and keeps the best one. Its selection method, Behavior Judge (BJudge), writes a short account of what each attempt did and compares those accounts.[11][4] Agent S4 adds a meta-agent, a second agent that studies the task agent's attempts on practice tasks and builds a library of interfaces and notes on how to use them. Different versions of the library compete, and the ones that do best on tasks the meta-agent has not seen are developed further.[5]
Benchmarks
[edit]On OSWorld, the first Agent S paper reported a success rate 9.37 percentage points above the baseline it was compared with.[1] Synced reported OSWorld scores of 20.6% for Agent S and 48.8% for Agent S2.[4] It gave 69.9% for Agent S3 at release, against a human score of about 72%.[4] The current version of the Agent S3 paper reports 72.6% when several attempts are run, against a human score of 72.36%.[11]
The Agent S2 paper also reported results on WindowsAgentArena and AndroidWorld, with relative improvements of 52.8% and 16.52% over the previous best methods.[8] The Agent S4 paper reports a binary score of 21.3% on OSWorld V2 using the open-weight Qwen3.8-27B model.[5]
Reception
[edit]The Decoder described Agent S as an attempt to take over routine office work such as data entry and scheduling.[2] It also reported the developers' own finding that about 40 percent of failures came from weak task planning or from matching commands to the wrong screen elements.[2] Wired wrote about Agent S2's use of more than one AI model.[3] Synced covered Agent S3 at length. It named inconsistent results from one run to the next as a main obstacle for computer-use agents, and explained how running attempts in parallel reduces it.[4]
Other researchers have used Agent S in their own work. The authors of OSWorld-Human, a benchmark that measures how efficiently computer-use agents work, studied Agent S2's runs on 39 OSWorld tasks to find where delays come from.[14] As of 2 November 2026, the GitHub repository had about 12,850 stars and 1,500 forks.[6]
References
[edit]- 1 2 3 4 5 Agashe, Saaket; Han, Jiuzhou; Gan, Shuyu; Yang, Jiachen; Li, Ang; Wang, Xin Eric (10 October 2024). "Agent S: An Open Agentic Framework that Uses Computers Like a Human". arXiv:2410.08164 [cs.AI].
- 1 2 3 4 5 Kemper, Jonathan (22 October 2024). "Agent S AI system automates complex computer tasks". The Decoder. Retrieved 2 November 2026.
- 1 2 3 Knight, Will (16 April 2025). "Meet the AI Agent With Multiple Personalities". Wired. Retrieved 2 November 2026.
- 1 2 3 4 5 6 机器之心编辑部 (6 October 2025). "你敢信？GPT-5的电脑操作水平只比人类低2%了" [Can you believe it? GPT-5's computer operation level is only 2% below humans]. 机器之心 (Synced) (in Chinese). Retrieved 2 November 2026 – via The Paper.
- 1 2 3 4 Agashe, Saaket; Yang, Yuzhe; Gonzalez-Pumariega, Gonzalo; Koduru, Bhuvan Chandra; Li, Chih-Lun; Yang, Jiachen; Li, Ang; Wang, Xin Eric (20 October 2026). "Evolving Agent Interfaces". arXiv:2610.28601 [cs.AI].
- 1 2 "simular-ai/Agent-S". GitHub. Retrieved 2 November 2026.
- ↑ Agashe, Saaket; Han, Jiuzhou; Gan, Shuyu; Yang, Jiachen; Li, Ang; Wang, Xin Eric (2025). Agent S: An Open Agentic Framework that Uses Computers Like a Human. International Conference on Learning Representations (ICLR 2025). OpenReview.
- 1 2 3 Agashe, Saaket; Wong, Kyle; Tu, Vincent; Yang, Jiachen; Li, Ang; Wang, Xin Eric (1 April 2025). "Agent S2: A Compositional Generalist-Specialist Framework for Computer Use Agents". arXiv:2504.00906 [cs.AI].
- ↑ Agashe, Saaket; Wong, Kyle; Tu, Vincent; Yang, Jiachen; Li, Ang; Wang, Xin Eric (2025). Agent S2: A Compositional Generalist-Specialist Framework for Computer Use Agents. Conference on Language Modeling (COLM 2025). OpenReview.
- 1 2 Claburn, Thomas (15 July 2025). "Former Google DeepMind engineer behind Simular says other AI agents are doing it wrong". The Register. Retrieved 2 November 2026.
- 1 2 3 Gonzalez-Pumariega, Gonzalo; Tu, Vincent; Lee, Chih-Lun; Yang, Jiachen; Li, Ang; Wang, Xin Eric (2 October 2025). "Scaling Agents for Computer Use". arXiv:2510.02250 [cs.AI].
- ↑ Gonzalez-Pumariega, Gonzalo; Tu, Vincent; Lee, Chih-Lun; Yang, Jiachen; Li, Ang; Wang, Xin Eric (2026). "Scaling Agents for Computer Use". Transactions on Machine Learning Research.
- ↑ Tang, Fei; Xu, Haolei; Zhang, Hang; et al. (27 March 2025). "A Survey on (M)LLM-Based GUI Agents". arXiv:2504.13865 [cs.HC].
- ↑ Abhyankar, Reyna; Qi, Qi; Zhang, Yiying (19 June 2025). "OSWorld-Human: Benchmarking the Efficiency of Computer-Use Agents". arXiv:2506.16042 [cs.AI].
External links
[edit]- Agent S on GitHub
Category:Free software programmed in Python Category:2024 software
