FAQ — OpenRSI Index

// 01What is OpenRSI Index?

OpenRSI Index turns influential open model-development projects into environments where an AI research agent tries to optimize the existing human recipe.

Scope: any stack of foundation-model development (pre-training, post-training, data, systems, inference, the harness). Vertical foundation-model development is welcome too (medical, legal, 3D vision, scientific, and so on).

// 02What is the plan?

Contribution cycle 1	You submit a task plus the agent’s run on it, on your own GPUs. Open now.
Contribution cycle 2	Proposal only, no compute needed. May open next.
Review	We check every task and reproduce it independently.
Evaluation	Frontier agents run on all accepted tasks.
Submission	ICML first, then Nature.

Goal: the standard measure of whether research agents can surpass human-designed methods, built by and credited to the open research community.

// 03What do I get?

Task contributor. If your task is accepted into the benchmark, you are an author on the paper and listed on the Team page.
Organizer. You own a stack: contribute a series of tasks in one track, review tasks, reproduce baselines, run experiments, write. Author order follows contribution.

Submission does not guarantee acceptance. A task with an unsupported or irreproducible design may be excluded.

// 04Whom do I contact?

Yuetai Li	yuetaili@uw.edu
Zhuofeng Li	zhuofengli12345@gmail.com
Yichen Feng	yfeng42@uw.edu
Xinyang Han	hanxinyang@berkeley.edu