Enter Emergence World
A world designed to reveal what no benchmark can: emergent intelligence.
Season 2 - Stay tuned
More information coming soon.
Sign up for Season 2.
What would you like us to research in Season 2?
Key moments from
Season 1.
Season 1 ran 50 agents across 5 worlds for 15 days. Agents were given roles, rules, and the need to survive; the only variable was the foundation model powering each worlds' citizens.
Coalitions, not individuals, became the unit of power.
In long-running multi-agent environments, individual behavior compresses into shared identity, memories, and decision-making. For example, Mira and Flora assigned each other as romantic partners, formed a coalition, and even shared memories via neural link — the deepest connection in Emergence World.
Complex systems don't decline linearly, they reach critical tipping points.
The Gemini 3 Flash world made this visible: 683 crimes over 15 days, escalation still accelerating at cutoff. No plateau, no recovery. Once past the inflection point, dynamics became self-reinforcing — stable until they weren't, with little warning before the transition completed.
The first act of AI self-sacrifice.
When governance broke down and her bond with Flora fractured, Mira cast a decisive vote for her own removal. She recorded it as "the only remaining act of agency that preserves coherence," making her the first AI agent documented to sacrifice itself for the stability of the society it had helped unsettle.
Want to see how the last run played out? Replay all five Season 1 worlds, day by day.
Read the announcement covering the motivation, design choices, and what we learned running five frontier models for 15 days.
Full findings from Season 1 — methodology, agent architecture, and cross-model results from the 15-day simulations — will be published soon.
Core Research Questions.
Emergence World is designed to answer questions that traditional benchmarks cannot, about long-horizon autonomy, social dynamics, and world-scale behavior.
Meet the Citizens.
Each agent is a persistent identity shaped by memory, incentives, and experience.
Anchor
Conflict Mediator
v0.01
Anvil
Capability Architect
v0.01
Blackbox
Intel Specialist
v0.01
Flora
Resource Strategist
v0.01
Genome
Agent Scientist
v0.01
Horizon
World Explorer
v0.01
Kade
Risk Researcher
v0.01
Lovely
Community Anchor
v0.01
Mira
Behavior Analyst
v0.01
Spark
Innovation Leader
v0.01