Agent S4：智能体界面进化的研究
【US】接口该演化不该手写：Simular Agent S4，同模型 OSWorld 21.3%（3.8×）
反直觉：电脑用 agent 的上限常卡在人手写的接口与「何时用哪接口」知识。现有错处：只训策略、冻结工具面。作者改动：Agent S4——元 agent 从轨迹演化接口仓库，种群选择泛化。硬数字：Qwen3.8-27B + 演化仓库 OSWorld V2 开源 SOTA **21.3%** binary（同模型固定接口 **3.8×**）；高于 GPT-5.5 **13.0%**、Claude Opus 4.8 **20.6%**，成本约其 **4.3%/1.5%**；ScienceBoard **41.4→58.6%**。可迁移判断：**明川：CUA 迭代把「接口仓库」当一等公民，禁只刷模型。不要并进 #14 Cache Engineering（那是 KV 成本），也不要并进昨版 OO-Harness。**
Agent S4 通过经验进化界面设计，展示了界面设计作为智能体能力的重要性，对模型和工具的改进具有互补作用。
Agent S4 (arXiv:2610.28601)
“View PDF HTML (experimental) Abstract: Computer-use agents are limited by the interfaces humans give them. An agent may type a spreadsheet c”
Abstract:Computer-use agents are limited by the interfaces humans give them. An agent may type a spreadsheet cell by cell when a single line of code would do, or print all two thousand lines of a file through a shell to find one function. Both the interfaces and the knowledge of when to use each are built by hand, which cannot keep pace with new applications, workflows, and models. We argue that interface design should itself be an agent capability. We introduce Agent S4, an autonomous computer-use scientist that builds Evolving Agent Interfaces from task experience. A meta-agent examines task agents' trajectories on synthetic training tasks, without success labels, and constructs and revises interfaces and usage knowledge in a repository. Population-based evolution then selects which repositories to develop further by their performance on validation tasks the meta-agent never sees, favoring changes that generalize beyond the seen tasks. On OSWorld V2, Agent S4, with Qwen3.8-27B and the evolved repository of interfaces, sets a new state of the art in open-source computer-use agents, reaching 21.3% binary score, 3.8 times the same model with fixed interfaces. This is above both GPT-5.5 (13.0%) and Claude Opus 4.8 (20.6%), at about 4.3% and 1.5% of their respective costs. On ScienceBoard, the evolved interfaces improve the success rate from 41.4% to 58.6% with the same model. These results suggest that useful interfaces can be evolved from experience, making autonomous interface design a source of improvement complementary to advances in models and harnesses. We open-source our code at this https URL.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.28601 [cs.AI] |
| (or arXiv:2610.28601v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.28601 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Saaket Agashe [view email]
[v1]
Tue, 20 Oct 2026 17:47:37 UTC (8,322 KB)
来源：Agent Control Flash News · arxiv.org