Simular's Agent S4 evolves computer-use interfaces, lifting Qwen3.8-27B to 21.3% on OSWorld V2
Simular introduces Agent S4, which does not solve tasks directly but evolves a repository of computer interfaces via population-based evolution while the task agent's model and harness stay fixed. A meta-agent selects parent repositories, examines their trajectories on synthetic training tasks without success labels, writes mutated or crossed-over repositories, and evaluates them on validation tasks. With the evolved repository, Qwen3.8-27B reaches 21.3% binary success on the 108-task OSWorld V2 benchmark, versus 5.56% for the fixed-interface Qwen CUA. On ScienceBoard, the evolved repository solves 99 of 169 tasks versus 93 for the seed.
One primary source; other links are context, not independent corroboration.
- Primary sourceX discovery · automated brief
- Discovery linkX
- Linked in sourcePaper
- Linked in sourceAgent S repository
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Open full article view →- Linked article+30
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
Simular introduces Agent S4, which does not solve tasks directly but evolves a repository of computer interfaces via population-based evolution while the task agent's model and harness stay fixed. A meta-agent selects parent repositories, examines their trajectories on synthetic training tasks without success labels, writes mutated or crossed-over repositories, and evaluates them on validation tasks. With the evolved repository, Qwen3.8-27B reaches 21.3% binary success on the 108-task OSWorld V2 benchmark, versus 5.56% for the fixed-interface Qwen CUA. On ScienceBoard, the evolved repository solves 99 of 169 tasks versus 93 for the seed.
Key findings from the stored brief
- With the evolved repository, Qwen3.8-27B reaches 21.3% binary success on OSWorld V2, versus 5.56% for Qwen CUA with fixed interfaces.
- The meta-agent learns only from synthetic training tasks with no success labels; it never sees benchmark tasks or validation trajectories.
- Running with a scratch VM built three times as many new interfaces and gave the best partial score; the best binary success came from the run without it.
- On ScienceBoard, the evolved repository solves 99 of 169 tasks, compared with 93 for the seed repository.
Source claims not independently checked; results are self-reported by the company on its own benchmark runs.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
The work treats interfaces as a separate axis of improvement from the model and harness. If the reported gains hold, open models could gain substantially from better interfaces alone, and repositories could keep evolving as models and applications change.
Limits and open questions
Source claims not independently checked; results are self-reported by the company on its own benchmark runs.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Ask about the source, add context, or point readers to a relevant follow-up.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.