# Three Roles, One Model: Role Orchestration at Inference Time to Close the Performance Gap Between Small and Large Agents

arXiv: https://arxiv.org/abs/2604.11465 · HTML: https://mlsanity.com/paper/2604.11465/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use.
- Builds on 29 of its 34 references that are in the map.
- First posted on arXiv 13 Apr 2026.

## Placed in

- LLM agents and tool use / [General LLM agent frameworks and benchmarks](https://mlsanity.com/agents/general-llm-agent-frameworks-and-benchmarks/)

## Builds on (most cited first, 29 of 29)

- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://mlsanity.com/paper/2203.11171/) (2022)
- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [Toolformer: Language Models Can Teach Themselves to Use Tools](https://mlsanity.com/paper/2302.04761/) (2023)
- [Self-Refine: Iterative Refinement with Self-Feedback](https://mlsanity.com/paper/2303.17651/) (2023)
- [Tree of Thoughts: Deliberate Problem Solving with Large Language Models](https://mlsanity.com/paper/2305.10601/) (2023)
- [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024)
- [Gorilla: Large Language Model Connected with Massive APIs](https://mlsanity.com/paper/2305.15334/) (2023)
- [Teaching Large Language Models to Self-Debug](https://mlsanity.com/paper/2304.05128/) (2023)
- [CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing](https://mlsanity.com/paper/2305.11738/) (2023)
- [API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs](https://mlsanity.com/paper/2304.08244/) (2023)
- [LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compression](https://mlsanity.com/paper/2310.06839/) (2023)
- [AppWorld: A Controllable World of Apps and People for Benchmarking Interactive Coding Agents](https://mlsanity.com/paper/2407.18901/) (2024)
- [Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models](https://mlsanity.com/paper/2510.04618/) (2025)
- [FireAct: Toward Language Agent Fine-tuning](https://mlsanity.com/paper/2310.05915/) (2023)
- [A Survey of Reinforcement Learning for Large Reasoning Models](https://mlsanity.com/paper/2509.08827/) (2025)
- [Dynamic Cheatsheet: Test-Time Learning with Adaptive Memory](https://mlsanity.com/paper/2504.07952/) (2025)
- [Debug like a Human: A Large Language Model Debugger via Verifying Runtime Execution Step-by-step](https://mlsanity.com/paper/2402.16906/) (2024)
- [Reinforcement Learning for Long-Horizon Interactive LLM Agents](https://mlsanity.com/paper/2502.01600/) (2025)
- [Recursive Language Models](https://mlsanity.com/paper/2512.24601/) (2025)
- [Reinforcement Learning for Self-Improving Agent with Skill Library](https://mlsanity.com/paper/2512.17102/) (2025)
- [Tuning Language Models by Proxy](https://mlsanity.com/paper/2401.08565/) (2024)
- [AgentEvolver: Towards Efficient Self-Evolving Agent System](https://mlsanity.com/paper/2511.10395/) (2025)
- [Remember Me, Refine Me: A Dynamic Procedural Memory Framework for Experience-Driven Agent Evolution](https://mlsanity.com/paper/2512.10696/) (2025)
- [Meta-RL Induces Exploration in Language Agents](https://mlsanity.com/paper/2512.16848/) (2025)
- [SPIRAL: Symbolic LLM Planning via Grounded and Reflective Search](https://mlsanity.com/paper/2512.23167/) (2025)
- [Enhancing LLM Planning Capabilities through Intrinsic Self-Critique](https://mlsanity.com/paper/2512.24103/) (2025)

## Abstract (arXiv)

Large language model (LLM) agents show promise on realistic tool-use tasks, but deploying capable agents on modest hardware remains challenging. We study whether inference-time scaffolding alone, without any additional training compute, can improve the performance of a small model in complex multi-step environments. Operating on a single 24GB GPU, we evaluate Qwen3-8B on the AppWorld benchmark under both full-precision and 4-bit quantized configurations. Without any intervention, the raw model achieves just 5.4% (FP16) and 3.0% (AWQ) task goal completion. Guided by a systematic failure mode analysis, we introduce a three-tier inference scaffolding pipeline that deploys the same frozen model in three distinct roles: (1) a summarization model that preserves critical artifacts (tokens, credentials, API responses) while compressing dialogue history; (2) the main agent model that reasons over the compressed context; and (3) an isolated correction model that reviews and revises the agent's code output without access to conversation history, breaking repetitive failure loops. Applied to the same unmodified model, this scaffolding yields 8.9% (FP16) and 5.9% (AWQ) task goal completion, roughly doubling performance in both settings, with particularly strong gains on difficulty-1 tasks (15.8% to 26.3% FP16; 5.3% to 14.0% AWQ). On full-precision inference, our scaffolded 8B model surpasses DeepSeek-Coder 33B Instruct (7.1%) from the original AppWorld evaluation, demonstrating that structured inference-time interventions can make small models competitive with systems 4 times their size. We formalize the approach as a scaffolded policy over a frozen base model, three invocations of the same weights with different conditioning, drawing connections to test-time compute scaling and action-space shaping in reinforcement learning.