Aviary: training language agents on challenging scientific tasks
Authors:
Preprint
Abstract
Solving complex real-world tasks requires cycles of actions and observations. This is particularly true in science, where tasks require many cycles of analysis, tool use, and experimentation. Language agents are promising for automating intellectual tasks in science because they can interact with tools via natural language or code. Yet their flexibility creates conceptual and practical challenges for software implementations, since agents may comprise non-standard components such as internal reasoning, planning, tool usage, as well as the inherent stochasticity of temperature-sampled language models. Here, we introduce Aviary, an extensible gymnasium for language agents. We formalize agents as policies solving language-grounded partially observable Markov decision processes, which we term language decision processes. We then implement five environments, including three challenging scientific environments: (1) manipulating DNA constructs for molecular cloning, (2) answering research questions by accessing scientific literature, and (3) engineering protein stability. These environments were selected for their focus on multi-step reasoning and their relevance to contemporary biology research. Finally, with online training and scaling inference-time compute, we show that language agents backed by open-source, non-frontier LLMs can match and exceed both frontier LLM agents and human experts on multiple tasks at up to 100x lower inference cost.Keywords
All Related Versions
This publication has 28 references indexed in Scilit:
- The rise and potential of large language model based agents: a surveyScience China Information Sciences, 2025
- Tool learning with large language models: a surveyFrontiers of Computer Science, 2025
- Symbolic learning enables self-evolving agentsAI Open, 2025
- ProtAgents: protein discovery via large language model multi-agent collaborations combining physics and machine learningDigital Discovery, 2024
- A new age in protein design empowered by deep learningCell Systems, 2023
- Illuminating protein space with a programmable generative modelNature, 2023
- Deep Learning in Protein Structural Modeling and DesignPatterns, 2020
- Computational Modeling of Protein Stability: Quantitative Analysis Reveals Solutions to Pervasive ProblemsStructure, 2020
- MMseqs2 enables sensitive protein sequence searching for the analysis of massive data setsNature Biotechnology, 2017
- Developability Index: A Rapid In Silico Tool for the Screening of Antibody Aggregation PropensityJournal of Pharmaceutical Sciences, 2012