# Agent S4 builds its own tools, and a small open model beats the giants at using a computer

URL: https://ai-pulse.aglaganova.com/blog/agent-s4-builds-its-own-tools-and-a-small-open-model-beats-the-giants-at-using-a-computer
Published: 2026-10-22
Author: Roman Aglaganov, Full-stack developer
Category: Research
Region: North America
Tags: agents, open-weights

> Instead of a bigger model, Simular’s researchers gave a 27-billion-parameter open model better tools, designed by another AI. On a hard computer-use test it edged past Claude Opus 4.8 at a tiny fraction of the cost.

![Illustration: a black node at the centre of a ring of grey tool-like shapes, one of them red, linked by thin lines](https://ai-pulse.aglaganova.com/api/media/file/agent-s4-1.png)

AI agents that use a computer are only as good as the controls they are given. A new paper from Simular, posted on 20 October, argues that those controls should be designed by AI too, and shows that doing so lets a small open model compete with the biggest commercial ones.

## The problem

The authors give two examples of how agents waste effort: typing a spreadsheet cell by cell when one line of code would do, or printing two thousand lines of a file to find one function. The shortcuts, and the knowledge of when to use them, are usually built by hand, which cannot keep up with new apps and models.

## How Agent S4 works

Agent S4 adds a “meta-agent” that watches ordinary task agents work through practice tasks. It does not know whether they succeeded. It builds and revises a library of interfaces, small tools and notes on when to use each, from what it sees. A process modelled on evolution then keeps the libraries that perform best on validation tasks the meta-agent never sees, which favours changes that generalise.

## The results

On OSWorld V2, a benchmark of real computer tasks, Qwen3.8-27B with an evolved library scored 21.3%, 3.8 times the same model with fixed interfaces. The paper puts that above GPT-5.5 (13.0%) and Claude Opus 4.8 (20.6%), at about 4.3% and 1.5% of their costs. On ScienceBoard, a test of scientific software, success rose from 41.4% to 58.6% with the same model. These are the authors’ own figures, and the code is open source.

The low absolute scores are a reminder of how hard these tasks still are: even the best system in the comparison fails most of them.

## Why it matters

Most progress in agents has come from bigger models. This work suggests a second lever: better tools, built from experience, which an organisation could improve without training a new model. It also helps open models, which can run on a company’s own hardware.

## Sources

- [arXiv: Evolving Agent Interfaces (Agashe et al., 2026)](https://arxiv.org/abs/2610.28601)
- [GitHub: simular-ai/Agent-S](https://github.com/simular-ai/Agent-S)