This work presents 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.
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.
Introducing Relay-Bench, an unsaturated, holistic, text-only benchmark that measures LLMs'ability to complete an assortment of tasks from distinct domains in a single prompt. The leading model, GPT-5.5 (xHigh), scores 43.3%. The test set entirely consists of composite problems: groups of single-domain subproblems that are strung together into challenges that require reasoning across multiple domains in combination. Many of these problems then have layers of complexity added through prompt encoding and deliberate context bloat. Domains tested include visual reasoning, coding, math, information extraction (with a focus on web search), problem-solving, general knowledge, and data analysis. No restrictions are imposed outside of the model harness, and models are explicitly encouraged to leverage code-execution, web searches, and all available tools. All problems are composed of two to thirteen subproblems and do not require multi-modal input or output.
In this paper, we develop a multi-agent pipeline-based approach for solving competitive programming problems via Large Language Models (LLMs). Specifically, we analyze the Writer-Reviewer pipeline where the Writer Agent produces Python solutions and the Reviewer Agent gives static natural language feedback. Experiments on a 96-problem AtCoder subset of LiveCodeBench involve comparing twelve pipeline setups which employ three different models (GPT-OSS-20B, Qwen3-Coder-30B-A3B-Instruct, and Qwen2.5-Coder-7B-Instruct) in various agent roles. Within this experimental setup, the review process increases the performance of GPT-OSS-20B from 87.5% to 91.7% Pass@1, with the bootstrap 95% confidence intervals overlapping, while the performance of Qwen3-Coder-30B-A3B-Instruct does not improve and the performance of Qwen2.5-Coder-7B-Instruct improves marginally from 0% to 1.0%. This may indicate that the static iterative feedback mechanism helps to further improve the performance of a strong Writer but not the performance of a weak Writer. A diagnostic audit suggests that the very low Qwen2.5-Coder-7B-Instruct scores mainly reflect structured-output compliance failures in our setup rather than standalone coding capability. Moreover, performance varies more when changing the Writer Agent than when changing the Reviewer Agent, with differences of up to 88.6 percentage points across Writers and up to 17.7 percentage points across Reviewers.
Temel Kaan Ekiz, M. Z. Konyar· ADBA Computer Science· 0 citations
LLM agents execute heterogeneous sequences of model calls within a single task: some invocations require careful reasoning, while others are structured steps such as formatting or tool-argument construction. Prior routing methods exploit this asymmetry by assigning easy invocations to a cheaper small model and difficult ones to a large model. Such policies reduce inference cost, but they leave the small model's capability unchanged, so attainable savings remain bounded by the work the student can already solve. MERA instead improves the small model itself, using a single model invocation as the unit of adaptation. In each cycle, MERA replays failed student invocations to obtain execution-verified teacher demonstrations, distills recurring procedures into an iteratively updated SkillBook, and fine-tunes a student LoRA adapter via supervised learning and optional GRPO. Routing serves as supporting machinery for deployment: the improved student is served behind a cost-calibrated router with verifier-backed fallback, and a candidate SkillBook, adapter, or router is admitted only when joint replay preserves task quality. Empirically, four-cycle adaptation raises Qwen2.5-Coder-1.5B from 28.7% to 49.7% pass on held-out HumanEval+MBPP. Under verifier-backed fallback, the deployed policy retains 88.3% pass at 60.8% of always-Luna cost. On TAU-2, a fine-tuned Qwen3.5-2B improves from 14/35 to 18/35 and matches an unadapted 4B model. These results indicate that verifier-backed multi-cycle adaptation can increase small-model capability, rather than only routing around a fixed student.
Yuhang Yao, Zeyu Wang, Wanyi Chen et al.· 0 citations
A ScholForge - hierarchical multi-agent LLM system that tackles end-to-end software engineering research end-to-end, with a natural-language research goal, and produces working code, experimental results, and a compiled manuscript without human intervention.
A. Stepin, Boris Tolstokulakov, V. Kulikov et al.· SIGSOFT FSE Companion· 0 citations
The point of the project is less to compete with Siri or Alexa than to show that a legible, few-thousand-line assistant is enough to open apps, type into them, run searches, coordinate concurrent actions, recover from single-step failures, and be driven entirely from a phone in another room -- without the LLM ever touching the keyboard.
ReASearch is presented, a unified framework for reasoning-driven optimization in which the agent autonomously decides what to evaluate, how to diagnose failures, which edits to make, and when to verify or restart.