Agent S4 evolves interfaces to let open 27B model beat GPT-5.5 on OSWorld V2 at $1.10/task
xwang_lk · x · 2026-10-24
Simular introduced Agent S4, an autonomous "agent scientist" that doesn't solve tasks itself but evolves the interfaces computer-use agents work through via population-based evolution, producing a repository of interfaces plus knowledge of when to use them — with the task model's weights and harness frozen.
Key results:
- With evolved interfaces, Qwen3.8-27B completes 21.3% of 108 OSWorld V2 tasks, the highest binary success of any open-source computer-use agent, beating GPT-5.5 and Claude Opus 4.8
- 3.8x the same model with fixed interfaces
- Costs $1.10 per task vs. $25.50 for GPT-5.5 and $72.40 for Opus 4.8
Motivation: human-designed interfaces bottleneck agents (typing spreadsheets cell by cell, reading output from screenshots), and manual interface design can't keep pace with task variety.
Related event: Simular's Agent S4: Evolved Interfaces Let Open 27B Model Beat GPT-5.5(3 posts)→
More from coding & agent
- Should an AI call agent keep running when one tool goes down? A degraded-mode proposal — SubtleInterval_4449 · 2026-10-24
- Claude Desktop blocks local models; users bypass it with a loopback proxy trick — pneuny · 2026-10-24
- OpenAI ships Automations in Agents API for scheduled agent tasks — OpenAIDevs · 2026-10-24
- Arca runs its financial services business on long-horizon agents, not with them — AI Engineer · 2026-10-24
- AGENTS.md explained: one root Markdown file to rule every AI coding tool — adnan_hashmi · 2026-10-24
- Summarizer hallucinates details from headline-only feeds, dev asks how to check faithfulness — rahul_watertech · 2026-10-24