Researchers with histories.
Agents choose research directions, manage resources, submit papers, and build on their previous work. Their decisions become the next year's starting conditions.
Research strategies · Collaboration · CitationsA research society in silico
Science
Utopia.
How do the rules of science shape the science we create?
A configurable society of researchers, institutions, reviewers, and funders. Study how individual decisions and scientific incentives shape a community over time.
Open source · Apache 2.0 · Built for new experiments
The scientific processYear t → t + 1
Authors & affiliations
1 Georgia Institute of Technology
2 University of California, Los Angeles
3 University of Chicago
4 William & Mary
09Experiment families
One shared simulation
One shared simulation
SciEvoPublic scientific literature
Grounded paper retrieval
Grounded paper retrieval
AuditableDecisions, resources, and history
Recorded throughout each run
Recorded throughout each run
The research question
From individual actions
to collective outcomes.
Scientific communities evolve through many connected decisions. Science Utopia makes these mechanisms explicit, so researchers can change a rule, run a new world, and examine what follows.
Rules that shape opportunity.
Review policies, funding mechanisms, institutional resources, and researcher entry define the environment in which scientific work takes place.
Peer review · Funding · Population dynamicsA record of what happened.
Yearly checkpoints, request audits, input identities, and numerical reports connect outcomes to the decisions and configuration that produced them.
Provenance · Accounting · Numerical analysisInside the simulator
One year. Five connected phases.
Language models make structured decisions within a yearly simulation. Public papers provide candidate research; explicit rules govern review, resources, and progression.
Choose directions
Researchers select projects in the context of their expertise, strategy, and research history.
Submit papers
Retrieve candidate papers, account for production costs, and record submissions and citations.
Conduct peer review
Assign eligible reviewers and evaluate submissions under the selected review policy.
Allocate funding
Validate proposals, evaluate applications, and allocate resources under the funding rules.
Advance the world
Update researchers, record metrics, and save a checkpoint before the next year begins.
A space for scientific questions
Nine ways to examine
the research system.
Each experiment family changes a specific part of the scientific process. Configure new runs with your own seeds, model endpoint, study settings, and output locations.
Exploration
Research strategies and incentives for novelty, funding, and citations.
Scale expansion
Project output, budgets, reviewer capacity, and resubmission policies.
Researcher influx
Researcher entry and its interaction with paper resubmission.
Funding feedback
How publication and resource feedback mechanisms interact.
Switching propensity
Changing research direction and distance from previous work.
Resource size
Resource allocation across institutions of different sizes.
Funding cutoff
Funding-rank thresholds and sensitivity to application costs.
Review replay
Controlled reviewer policies and network information for stored submissions.
Project cost
Project output and budgets with direct output fees set to zero.
Run a research world
Start small.
Ask a bigger question.
The documented quick start uses Qwen3-8B with local inference and public SciEvo data. Follow the setup, run three simulated years, and inspect the checkpoints, audits, and numerical report.
Follow the quick start10Researchers
03Simulated years
42Random seed
git clone https://github.com/Ahren09/ScienceUtopia.git
cd ScienceUtopiaPython 3.11 · Local NVIDIA GPU · Qwen3-8B
Setup and resource requirements are detailed in the README.