Paper 2602.02660
MARS: Modular Agent with Reflective Search for Automated AI Research
- Published
- Feb 2026
- Research lab
- Google DeepMind
- Citations
- 12
- GitHub
- 39 stars
01 In brief
Summary
MARS (Modular Agent with Reflective Search) is a framework for automating AI research, specifically addressing the bottleneck of complex machine learning engineering (MLE) tasks.
It uses three pillars: Budget-Aware Planning via cost-constrained Monte Carlo Tree Search (MCTS) to balance performance with execution cost, Modular Construction through a 'Design-Decompose-Implement' pipeline for managing complex repositories, and Comparative Reflective Memory to solve credit assignment by analyzing solution differences.
On MLE-Bench, MARS achieves state-of-the-art performance among open-source frameworks under comparable settings, with the standard version reaching a 31.1% Gold Medal rate and the scaled MARS+ variant achieving 62.7% Any Medal rate.
Ablations validate each component, and qualitative analysis shows 63% of utilized lessons come from cross-branch transfer, demonstrating effective generalization.
The system also exhibits 'Aha!' moments, such as evolving from a lightweight residual network to an ensemble to achieve a silver medal on the iMet-2020-FGVC7 task.
02 From the paper
Abstract
A critical bottleneck in automating AI research is the execution of complex machine learning engineering (MLE) tasks. MLE differs from general software engineering due to computationally expensive evaluation (e.g., model training) and opaque performance attribution. Current LLM-based agents struggle here, often generating monolithic scripts that ignore execution costs and causal factors. We introduce MARS (Modular Agent with Reflective Search), a framework optimized for autonomous AI research. MARS relies on three pillars: (1) Budget-Aware Planning via cost-constrained Monte Carlo Tree Search (MCTS) to explicitly balance performance with execution expense; (2) Modular Construction, employing a "Design-Decompose-Implement" pipeline to manage complex research repositories; and (3) Comparative Reflective Memory, which addresses credit assignment by analyzing solution differences to distill high-signal insights. MARS achieves state-of-the-art performance among open-source frameworks on MLE-Bench under comparable settings, maintaining competitiveness with the global leaderboard's top methods. Furthermore, the system exhibits qualitative "Aha!" moments, where 63% of all utilized lessons originate from cross-branch transfer, demonstrating that the agent effectively generalizes insights across search paths.