ABSClaw: From Invented Worlds to Reality-Grounded Social Simulation
Organizations: Tianjin University · Zhongguancun Academy · Zhongguancun Institute of Artificial Intelligence · Beijing Normal University
Abstract
LLM-based artificial societies exhibit rich behavior but often lack scientific grounding: personas and environments are assumed, interaction networks are synthetic, and simulators are generated as free-form code, making results difficult to tie to real populations and sensitive to the backbone model. We argue that simulated worlds should be sampled from observed data, while simulation logic should be implemented by composing operations from a shared, task-independent vocabulary. In this paper, we introduce ABSClaw, a zero-code agent that compiles natural-language requests into executable, reality-grounded social simulations. ABSClaw grounds who acts, where they act, and with whom they interact in survey and census data, geospatial sources, and real networks. It expresses state transitions through 26 atomic operations. To support an open and extensible ecosystem, ABSClaw further allows capabilities to be added as lifecycle-scoped skills without modifying its core. Furthermore, we propose ABSBench, a benchmark of 193 real-world social-simulation tasks. On ABSBench, ABSClaw achieves the best overall performance among the evaluated social simulators. Three real-world case studies show that grounding personas, environments, and interaction networks each improves agreement with observed outcomes. With one runtime skill, ABSClaw executes 13 population-scale simulations, collectively involving 25 million agents. Together, these results demonstrate ABSClaw's real-world fidelity, extensibility, and scalability.
Figures & tables
| Category | Atomic operations |
|---|---|
| Attributes | set_attr , get |
| Entity lifecycle | create_agent , create_entity |
| Relations | create_edge , set_edge , get_edge , get_neighbors |
| Resources | transfer |
| Communication | send_message , read_messages , delete_message |
| Memory | add_memory , retrieve_memory |
| ABSClaw | Spec-to-Code | ||||||
|---|---|---|---|---|---|---|---|
| 95% CI | |||||||
| Opus 4.8 | 92.8 | 71.5 | 164.3 | 79.1 | 56.0 | 135.1 | +29.2 [20.4,38.0] |
| Qwen-A10B | 91.6 | 53.6 | 145.2 | 67.4 | 36.0 | 103.4 | +41.8 [33.0,50.9] |
| Qwen-27B | 96.8 | 59.1 | 155.9 | 57.5 | 28.7 | 86.2 | +69.6 [60.4,79.1] |
| GPT-5.5 | 90.1 | 78.5 | 168.6 | 80.7 | 56.6 | 137.3 | +31.2 [23.8,38.6] |
| Haiku 4.5 | 86.9 | 55.8 | 142.7 | 46.8 | 23.1 | 69.9 | +72.8 [62.3,83.3] |
| Held-out scenario | Agents | |
| Disaster insurance and institutional trust | 1.5M | 167 |
| Agricultural pest-response coordination | 1.0M | 158 |
| Open-science practice adoption | 0.5M | 175 |
| Remote work and regional migration | 3.0M | 183 |
| Colorectal-screening uptake | 2.0M | 149 |
| Public-benefit access and take-up | 2.5M | 162 |
Large Language Model-Based Social Simulation
Large language model-based social simulation studies how language-model agents represent people and interact to produce social behavior.
See all 292 papers