Skip to content
NewA calmer, faster ZeroHour
StoryAI researchinfo

Agent S4 evolves computer-use interfaces, scoring 21.3% on OSWorld V2

2 sources · 2 articlesFirst reported Updated ()
What's new

October 20: the arXiv posting names github.com/simular-ai/agent-s and says evolution uses synthetic training tasks plus unseen validation tasks.

Merged summary · x-ai/grok-4.7 · rewritten as coverage arrivesSummaries by Vani AI

Agent S4 lifts Qwen3.8-27B to 21.3% on OSWorld V2, above cited GPT-5.5 and Claude Opus 4.8 scores, and open-sources the code.

A paper introduces Agent S4, a meta-agent that builds and revises computer-use interfaces from unlabeled task trajectories and selects repositories by performance on unseen validation tasks. On OSWorld V2, Qwen3.8-27B with the evolved repository reaches a 21.3% binary score, 3.8 times the same model with fixed interfaces, above the paper’s cited GPT-5.5 score of 13.0% and Claude Opus 4.8 score of 20.6%. The authors put that run’s cost at about 4.3% of GPT-5.5 and 1.5% of Claude Opus 4.8. On ScienceBoard, the same model rises from 41.4% to 58.6%. The code is open-sourced.

Key facts
  • Qwen3.8-27B with evolved interfaces scores 21.3% binary on OSWorld V2.
  • That is 3.8 times the same model with fixed interfaces.
  • Cited comparisons: GPT-5.5 at 13.0% and Claude Opus 4.8 at 20.6%.
  • Authors put cost at about 4.3% of GPT-5.5 and 1.5% of Claude Opus 4.8.
  • ScienceBoard success for the same model rose from 41.4% to 58.6%.
  • Code is at github.com/simular-ai/agent-s.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

What each outlet adds

beyond the shared facts
  • Hugging Face daily papersReports the meta-agent revises interfaces without success labels.
  • arXiv cs.AI / cs.LG / cs.CLGives the code path and synthetic-plus-unseen task split.
Products
Agent S4OSWorld V2ScienceBoard
AI models
Qwen3.8-27BGPT-5.5Claude Opus 4.8

Coverage

oldest first · one row per article · UTC
  1. · 5d ago Hugging Face daily papersFirst report

    Agent S4 evolves interfaces and leads open-source OSWorld V2

    Qwen3.8-27B with evolved interfaces scores 21.3%, above GPT-5.5 and Claude Opus 4.8 at far lower cost.

  2. · 5d ago arXiv cs.AI / cs.LG / cs.CL

    Agent S4 evolves interfaces and leads open-source OSWorld

    With Qwen3.8-27B it scores 21.3% on OSWorld V2, 3.8 times the same model with fixed interfaces.