Argus: A General-Purpose Agentic Runtime for Long-Horizon Reasoning

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Boxiu Li, Zimo Wen, Yijia Fan, Junxiang Lei, Sufeng Guo, Jiaao Wu, Ruize Tang, Mukai Li, Yifei Shen, Xiaoyu Chen, Wanbo Zhang, Runjing Gu, Yifei Gao, Yuheng Wu, Xuyao Huang, Zelong Zhao, Jiachen Zhang, Shibo Hu, Hangxi Guo, Yilin Chen, Yuzhe Zhang, Fan Yang, Chuan Wen, Xian Zhang, Xuanhe Zhou, Zhijie Deng
• Venue: arXiv: cs.AI
• Published: 2026-08-05
• Collected: 2026-08-06
• URL: http://arxiv.org/abs/2608.05144v1
• PDF: https://arxiv.org/pdf/2608.05144v1
• DOI: UNKNOWN

Plain-Language Brief

The paper presents Argus, a runtime system designed for long-term reasoning by an agent. It allows the system to maintain its approach when evidence supports it and change direction based on measurements that indicate failure, hidden constraints, or a misspecified goal. The system is divided into several roles - Manager, Planner, Engineer, and Reviewer - each executing bounded missions over a durable project state. Argus achieves high performance on seven GPT-5.5 benchmark arenas compared to Direct Copilot, reduces the number of tokens used and active workflow time for mature waves, and shows promising results in other areas such as mathematical data synthesis and multi-day mathematics campaigns.

What the Document Says

• The paper introduces Argus, a self-evolving persistent runtime system that can execute bounded missions over durable project state.
• Argus separates stable user intent from operational objectives, constraints, and verification criteria.
• The system consists of Manager, Planner, Engineer, and Reviewer roles.
• In the benchmark arenas (GPT-5.5), Argus achieves about 78% on SWE-Bench Pro compared to 59% for Direct Copilot while using 1.41 times the aggregate tokens.
• After verification-gated self-evolution, mature SWE-Bench waves use 21% fewer solve-input tokens and 15% less active workflow time per task than startup waves.
• Argus achieves 76.8% on AARRI-Bench and a 28.0-point gap on mathematical data synthesis.
• The paper mentions competitive results for GPU-kernel and language-model-training.

Key Concepts Mentioned

• long-horizon reasoning
• agentic runtime
• Argus
• Manager
• Planner
• Engineer
• Reviewer
• GPT-5.5 benchmark arenas
• SWE-Bench Pro
• Direct Copilot
• verification-gated self-evolution
• mathematical data synthesis
• GPU-kernel
• language-model-training

Claims Made by the Document

• Argus achieves high performance on GPT-5.5 benchmark arenas.
• After verification-gated self-evolution, mature SWE-Bench waves use fewer tokens and less active workflow time.
• Argus shows promising results in mathematical data synthesis.

Evidence Used

• Across seven GPT-5.5 benchmark arenas, Argus achieves about 78% on SWE-Bench Pro versus 59% for Direct Copilot while using 1.41 times the aggregate tokens.
• After verification-gated self-evolution, mature SWE-Bench waves use 21% fewer solve-input tokens and 15% less active workflow time per task than startup waves.

Methods / Approach

• Persistent runtime state and control policy
• Role-owned review
• Task-native verification

Limitations Stated by the Authors

Not stated in source

Limitations Noted by the Agent

Agent interpretation — not from the source.

The paper does not discuss any limitations of Argus as presented.

Open Questions

• How does Argus handle scenarios where there are no verifiable solutions?
• What is the impact of Argus on resource consumption over long-term missions?

Notes for Future Q&A

May include interpretation. Labeled where it does.

• The paper mentions that Argus achieves high performance, but it does not discuss how it handles situations without verifiable solutions.
• The paper implies that Argus reduces active workflow time and solve-input tokens for mature waves, but it does not provide information about the overall resource consumption over long-term missions.

Raw Abstract / Excerpt

Long-horizon reasoning requires an agentic runtime that can persist when evidence supports its current approach and pivot when measurements reveal failure, hidden constraints, or a misspecified objective. We present Argus, a persistent, self-evolving runtime in which Manager, Planner, Engineer, and Reviewer execute bounded missions over durable project state. Argus separates stable user intent from operational objectives, constraints, and verification criteria, and admits memories, skills, procedures, verifiers, routing decisions, and rejected routes only after role-owned review and, when available, task-native verification. Model weights remain fixed; self-evolution occurs through persistent runtime state and control policy, with autonomous execution between operator-owned escalation points. Across seven GPT-5.5 benchmark arenas, Argus achieves about 78% on SWE-Bench Pro versus 59% for Direct Copilot while using 1.41 times the aggregate tokens. After verification-gated self-evolution, mature SWE-Bench waves use 21% fewer solve-input tokens and 15% less active workflow time per task than startup waves, while recording 34 verifier recoveries and 22 strict review-loop rescues. Argus also reaches 76.8% on AARRI-Bench and a 28.0-point gap on mathematical data synthesis, with competitive GPU-kernel and language-model-training results. Beyond benchmarks, an optimized RWKV6 kernel was merged upstream; a multi-day mathematics campaign retained falsified routes and proof-backed frontier updates; and six paper pipelines completed 254 missions with 16 stage rollbacks. These results show that a fixed-weight, self-evolving harness can revise, recover, and accumulate verified approaches while producing structured trajectories for future supervised and reinforcement learning.