Harvard And Mit · depuis 4. August 2026 (arXiv-Preprint-Veröffentlichung) · 3× · vu le 11 août 2026
79
Momentum
MatrAIx is an open-source evaluation infrastructure developed by researchers from Harvard and MIT for testing AI systems and digital products with simulated users. Its core is "Persona 8B", a database of 8.3 billion persona records spanning 1,290 categorical dimensions, built from both synthetic generation and human-grounded profiles. Through the "MatrAIx Playground" environments (Survey, AI Chatbot, Web, App), developers can run persona-based AI agents against their own products; a public coreset of about 1 million personas and the source code are available via Hugging Face and GitHub respectively.
Historique du momentum
13.05.11.08.
Fonctionnalités
| Throughput/Latency | 18,189 evaluation trials completed across 8 tasks; 91.5% persona behavioral adherence (366/400 trials) |
| License | MIT License (GitHub repository MatrAIx-Persona-8B) |
| Platform | Web playground, Python package (pip install matraix), GitHub repository, Hugging Face dataset |
| Price | Free/open source (code & coreset dataset freely available); commercial API via pip install matraix, pricing details not publicly listed |
| Protocol Compatibility | Proprietary Python API (matraix.connect); four interaction environments: Survey, AI Chatbot, Web, App (incl. macOS/iOS) |
| Release Date | arXiv paper submitted August 4, 2026 |
| Supported Models/Providers | Persona agents powered by Claude Opus 4.8, GPT 5.5, and Claude Haiku 4.5; API keys configurable for Anthropic and OpenAI models |