aiqbench analysis
MARS introduces a specialized multi-agent framework where LLMs act as domain-specific experts (e.g., dynamic programming, graph theory) rather than relying on a single backbone model to select algorithms. By using retrieval-augmented generation to assemble a tailored team of specialists for each coding problem, the system significantly improves pass rates on competitive programming benchmarks while reducing computational costs compared to generic prompting methods. This approach addresses a key failure mode in current code-generation pipelines by explicitly separating algorithmic strategy selection from general coding execution.
MARS: Multi-Specialist LLM Relay System for Competitive Programming
arXiv:2608.23918v1 Announce Type: new Abstract: Large Language Models excel at code generation, yet competitive programming exposes a persistent failure mode: existing multi-agent pipelines distribute work over generic planner, coder, and debugger roles and delegate the choice of algorithmic technique to the backbone alone. We present MARS (Multi-Agent Relay of Specialized LLMs), a prompt-only framework in which each agent is a topic specialist---dynamic programming, graphs, strings, geometry, and so on---grounded by retrieval-augmented generation over an algorithm-theory corpus. Given a problem, retrieval selects a small team of relevant specialists; a starter writes an initial C++17 solution, and each subsequent turn runs the candidate against public examples in a sandbox, lets the active specialist keep, repair, or hand off the draft, and forwards a structured packet to the next specialist. A single infrastructure-fixer pass normalizes boilerplate at the end. On the CodeContests test split with Gemma 4, MARS reaches $0.624 \pm 0.006$ pass rate at $2.3$ recorded pipeline stages per task ($+14.4$ percentage points over direct prompting), closing most of the gap to CodeSIM ($0.731$) at $3.3{\times}$ lower wall-clock cost and substantially smaller variance in per-task token spend. The source code is available on GitHub: https://github.com/fckand/mars.
Abstract:Large Language Models excel at code generation, yet competitive programming exposes a persistent failure mode: existing multi-agent pipelines distribute work over generic planner, coder, and debugger roles and delegate the choice of algorithmic technique to the backbone alone. We present MARS (Multi-Agent Relay of Specialized LLMs), a prompt-only framework in which each agent is a topic specialist---dynamic programming, graphs, strings, geometry, and so on---grounded by retrieval-augmented generation over an algorithm-theory corpus. Given a problem, retrieval selects a small team of relevant specialists; a starter writes an initial C++17 solution, and each subsequent turn runs the candidate against public examples in a sandbox, lets the active specialist keep, repair, or hand off the draft, and forwards a structured packet to the next specialist. A single infrastructure-fixer pass normalizes boilerplate at the end. On the CodeContests test split with Gemma 4, MARS reaches $0.624 \pm 0.006$ pass rate at $2.3$ recorded pipeline stages per task ($+14.4$ percentage points over direct prompting), closing most of the gap to CodeSIM ($0.731$) at $3.3{\times}$ lower wall-clock cost and substantially smaller variance in per-task token spend. The source code is available on GitHub: this https URL.
Artificial Intelligence (cs.AI); Multiagent Systems (cs.MA); Programming Languages (cs.PL)
arXiv:2608.23918 [cs.AI]
(or arXiv:2608.23918v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2608.23918 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 290 words
Related coverage
GrandCode: Achieving Grandmaster Level in Competitive Programming via Agentic Reinforcement Learning
arXiv · Aug 7, 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · Oct 2, 2026
Top Photonic Quantum Computing Companies
Quantum Zeitgeist · May 27, 2026
AI and Quantum Computing Briefing
livetradingnews.com · Aug 4, 2026