Terminus-4B: Can a Smaller Model Replace Frontier LLMs at Agentic Execution Tasks?

• URL: http://arxiv.org/abs/2605.03195v1
• Date: Mon, 04 May 2026 22:24:24 GMT
• Title: Terminus-4B: Can a Smaller Model Replace Frontier LLMs at Agentic Execution Tasks?
• Abstract summary: Small language model (SLM) can achieve comparable performance to frontier models in the task of agentic terminal execution.<n>We present Terminus-4B, which is a post-trained Qwen3-4B model via Supervised Finetuning (SFT) and Reinforcement Learning (RL)<n>We find that Terminus-4B is able to reduce the token usage of the main agent by up to 30% compared to the No Subagent baseline.
• Score: 3.636948650519796
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Modern coding agents increasingly delegate specialized subtasks to subagents, which are smaller, focused agentic loops that handle narrow responsibilities like search, debugging or terminal execution. This architectural pattern keeps the main agent's context window clean by isolating verbose outputs (e.g. build logs, test results, etc.) within the subagent context. Typically when agents employ subagents for such tasks, they use frontier models as these subagents. In this paper, we investigate whether a finetuned small language model (SLM) can achieve comparable performance to frontier models in the task of agentic terminal execution. We present Terminus-4B, which is a post-trained Qwen3-4B model via Supervised Finetuning (SFT) and Reinforcement Learning (RL) using rubric-based LLM-as-judge reward, specifically for this task. In our extensive evaluation spanning various frontier models, training ablations and main agent configurations, we find that Terminus-4B is able to reduce the token usage of the main agent by up to ~30% compared to the No Subagent baseline with no impact to agent performance on benchmarks like SWE-Bench Pro and our internal SWE-Bench C# benchmark, which tends to be heavy in verbose execution tasks. Furthermore, Terminus-4B improves key metrics showing the main agent relying on the outputs of the subagent and doing fewer terminal execution tasks by itself. We see that our model not only closes the gap between the Vanilla Qwen model and frontier models like Claude Sonnet / Opus / GPT-5.3-Codex, but often even exceeds their performance.

Related papers

• Baselines Before Architecture: Evaluating Coding Agents for Autonomous Penetration Testing [0.0]
  This paper presents a controlled study on the 104-task XBOW benchmark using default coding CLI agents as plain-agent baselines.<n>We first run Codex, OpenCode, and Pi with the same GPT-5 model, budget, target interface, and scoring rule.<n>We then compare the default Codex scaffold with published MAPTA and PentestGPT V2 results under the closest available model matches.
  arXiv  Detail & Related papers  (2026-07-13T13:31:46Z)
• ClawArena-Team: Benchmarking Subagent Orchestration and Dynamic Workflows in Language-Model Agents [61.184954205350984]
  ClawArena-Team is a benchmark of 41 multi-turn, multimodal, multi-agent scenarios spanning 258 evaluation rounds and 72 staged updates.<n>All scoring is execution-based with no LLM judge: an overall score -- the Subagent-Management Score (SMS) -- multiplies task correctness by a least-privilege and modality-routing factor.
  arXiv  Detail & Related papers  (2026-06-30T06:06:08Z)
• Multi-Agent Computer Use [72.79887808312706]
  We argue that we should move towards evaluating and building multi-agent computer use (MACU) systems.<n>We propose a general multi-agent setup in which a manager model decomposes computer use tasks as a directed acyclic graph (DAG)<n>At each iteration, the manager dispatches parallel CUA subagents to carry out nodes on the ready frontier of the DAG.
  arXiv  Detail & Related papers  (2026-06-01T01:29:36Z)
• Latent Agents: A Post-Training Procedure for Internalized Multi-Agent Debate [16.68699018207298]
  Multi-agent debate has been shown to improve reasoning in large language models (LLMs)<n>We develop a framework that distills multi-agent debate into a single LLM through a two-stage fine-tuning pipeline.<n>Our internalized models match or exceed explicit multi-agent debate performance using up to 93% fewer tokens.
  arXiv  Detail & Related papers  (2026-04-27T18:06:03Z)
• AgentArk: Distilling Multi-Agent Intelligence into a Single LLM Agent [57.10083973844841]
  AgentArk is a novel framework to distill multi-agent dynamics into the weights of a single model.<n>We investigate three hierarchical distillation strategies across various models, tasks, scaling, and scenarios.<n>By shifting the burden of computation from inference to training, the distilled models preserve the efficiency of one agent while exhibiting strong reasoning and self-correction performance of multiple agents.
  arXiv  Detail & Related papers  (2026-02-03T19:18:28Z)
• ROMA: Recursive Open Meta-Agent Framework for Long-Horizon Multi-Agent Systems [25.131570054560353]
  Current agentic frameworks underperform on long-horizon tasks.<n>We introduce ROMA, a domain-agnostic framework that addresses these limitations.<n>We show that ROMA, combined with GEPA+, delivers leading system-level performance on reasoning and long-form generation benchmarks.
  arXiv  Detail & Related papers  (2026-02-02T09:20:59Z)
• Behavior Knowledge Merge in Reinforced Agentic Models [48.89546963456286]
  Reinforcement learning is central to post-training, particularly for agentic models that require specialized reasoning behaviors.<n>Existing merging methods are designed for supervised fine-tuning (SFT), and they are suboptimal to preserve task-specific capabilities on RL-trained agentic models.<n>We propose Reinforced Agent Merging (RAM), a distribution-aware merging framework explicitly designed for RL-trained agentic models.
  arXiv  Detail & Related papers  (2026-01-20T03:56:53Z)
• Stop Wasting Your Tokens: Towards Efficient Runtime Multi-Agent Systems [11.42175340352007]
  We introduce SupervisorAgent, a lightweight and modular framework for runtime, adaptive supervision.<n>SupervisorAgent intervenes at critical junctures to proactively correct errors, guide inefficient behaviors, and purify observations.<n>On the challenging GAIA benchmark, SupervisorAgent reduces the token consumption of the Smolagent framework by an average of 29.45% without compromising its success rate.
  arXiv  Detail & Related papers  (2025-10-30T15:12:59Z)
• Reinforcement Learning for Machine Learning Engineering Agents [52.03168614623642]
  We show that agents backed by weaker models that improve via reinforcement learning can outperform agents backed by much larger, but static models.<n>We propose duration- aware gradient updates in a distributed asynchronous RL framework to amplify high-cost but high-reward actions.<n>We also propose environment instrumentation to offer partial credit, distinguishing almost-correct programs from those that fail early.
  arXiv  Detail & Related papers  (2025-09-01T18:04:10Z)
• Visual Document Understanding and Question Answering: A Multi-Agent Collaboration Framework with Test-Time Scaling [83.78874399606379]
  We propose MACT, a Multi-Agent Collaboration framework with Test-Time scaling.<n>It comprises four distinct small-scale agents, with clearly defined roles and effective collaboration.<n>It shows superior performance with a smaller parameter scale without sacrificing the ability of general and mathematical tasks.
  arXiv  Detail & Related papers  (2025-08-05T12:52:09Z)
• Distilling LLM Agent into Small Models with Retrieval and Code Tools [65.73762766854192]
  Agent Distillation is a framework for transferring reasoning capability and task-solving behavior from large language models into small language models.<n>Our results show that sLMs as small as 0.5B, 1.5B, 3B parameters can achieve performance competitive with next-tier larger 1.5B, 3B, 7B models.
  arXiv  Detail & Related papers  (2025-05-23T08:20:15Z)
• Process Reward Models for LLM Agents: Practical Framework and Directions [10.986389591866617]
  We introduce Agent Process Reward Models (AgentPRM), a framework for training LLM agents to continually improve through interactions.<n>We propose InversePRM, which learns process rewards directly from demonstrations without explicit outcome supervision.<n>We evaluate on ALFWorld benchmark, show that small 3B models trained with AgentPRM and InversePRM outperform strong GPT-4o baselines.
  arXiv  Detail & Related papers  (2025-02-14T17:34:28Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.