Simular's Agent S4 Beats GPT-5.5 by Evolving Interfaces, Not Weights

Simular's Agent S4 shows that automatically evolving an agent's computer-use interface—without changing model weights or the agent framework—lifts an open-source 27B model from 5.6% to 21.3% on OSWorld V2, surpassing GPT-5.5.

2026-10-26 ~ 2026-10-26 · 2 related posts

• Simular's Agent S4 lifts a 27B open model to beat GPT-5.5 on OSWorld at $1.10 per task — xwang_lk · 2026-10-26
• Agent S4: a 27B model jumps 5.6% to 21.3% on OSWorld without any weight changes — krishnan · 2026-10-26