Open-Source Claude Science Rival Released: Zero Dependencies, MIT License, 30+ Research Skills Built In
Anthropic
Peking University and YuanKong AI have open-sourced OpenAI4S, a research agent replicating Claude Science with a zero-dependency core under the MIT license. It features Code-as-Action, a persistent kernel, host-RPC, and a no-fabrication policy, with over 30 built-in research skills.
OpenAI4S, developed by the Peking University & YuanKong AI Agent Joint Laboratory, was open-sourced on GitHub on July 6, 2026, under the MIT license. It independently reproduces the architecture of Anthropic's Claude Science, focusing on Code-as-Action where the agent generates Python or R code executed in a persistent kernel. The project includes a full research web app, dozens of built-in Skills, support for connecting own compute resources, and versioned research outputs. Key features include the ability to handle complex multi-step research tasks, a no-fabrication policy that prohibits fake data, and integration with real services like UniProt and RCSB for protein analysis. The project also supports GPU-based computations via SSH to remote servers for tasks like protein structure prediction using Protenix and mutation scoring with ESM. OpenAI4S aims to be an open research infrastructure, inviting contributions from the community.
- Abbreviations
- MIT = Massachusetts Institute of Technology — Массачусетский технологический институт
- RPC = Remote Procedure Call — удалённый вызов процедур
- GPU = Graphics Processing Unit — графический процессор
- ESM = Evolutionary Scale Modeling — эволюционное масштабное моделирование
Source: QbitAI 量子位 —
original
