# Reasoning as a Resource: Optimizing Fast and Slow Thinking in Code Generation Models

arXiv: https://arxiv.org/abs/2506.09396 · HTML: https://mlsanity.com/paper/2506.09396/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning.
- Builds on 22 of its 32 references that are in the map.
- First posted on arXiv 11 Jun 2025.

## Placed in

- LLM reasoning / [Reasoning for agents, planning and code](https://mlsanity.com/lr/reasoning-for-agents-planning-and-code/)

## Builds on (most cited first, 22 of 22)

- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://mlsanity.com/paper/2203.11171/) (2022)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [Code Llama: Open Foundation Models for Code](https://mlsanity.com/paper/2308.12950/) (2023)
- [StarCoder: may the source be with you!](https://mlsanity.com/paper/2305.06161/) (2023)
- [Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes](https://mlsanity.com/paper/2305.02301/) (2023)
- [StarCoder 2 and The Stack v2: The Next Generation](https://mlsanity.com/paper/2402.19173/) (2024)
- [AdaptThink: Reasoning Models Can Learn When to Think](https://mlsanity.com/paper/2505.13417/) (2025)
- [Verify-and-Edit: A Knowledge-Enhanced Chain-of-Thought Framework](https://mlsanity.com/paper/2305.03268/) (2023)
- [Structured Chain-of-Thought Prompting for Code Generation](https://mlsanity.com/paper/2305.06599/) (2023)
- [Split and Merge: Aligning Position Biases in LLM-based Evaluators](https://mlsanity.com/paper/2310.01432/) (2023)
- [SelfDefend: LLMs Can Defend Themselves against Jailbreaking in a Practical Manner](https://mlsanity.com/paper/2406.05498/) (2024)
- [RepoAudit: An Autonomous LLM-Agent for Repository-Level Code Auditing](https://mlsanity.com/paper/2501.18160/) (2025)
- [HumanEval Pro and MBPP Pro: Evaluating Large Language Models on Self-invoking Code Generation](https://mlsanity.com/paper/2412.21199/) (2024)
- [CCTEST: Testing and Repairing Code Completion Systems](https://mlsanity.com/paper/2208.08289/) (2022)
- [Refining Decompiled C Code with Large Language Models](https://mlsanity.com/paper/2310.06530/) (2023)
- [Fixing Large Language Models' Specification Misunderstanding for Better Code Generation](https://mlsanity.com/paper/2309.16120/) (2023)
- [Exploring Multi-Lingual Bias of Large Code Models in Code Generation](https://mlsanity.com/paper/2404.19368/) (2024)
- [API-guided Dataset Synthesis to Finetune Large Code Models](https://mlsanity.com/paper/2408.08343/) (2024)
- [Unleashing the Power of Compiler Intermediate Representation to Enhance Neural Program Embeddings](https://mlsanity.com/paper/2204.09191/) (2022)
- [Low-Cost and Comprehensive Non-textual Input Fuzzing with LLM-Synthesized Input Generators](https://mlsanity.com/paper/2501.19282/) (2025)
- [SPVR: syntax-to-prompt vulnerability repair based on large language models](https://mlsanity.com/paper/2405.04994/) (2024)

## Built on by (most cited first, 3 of 3)

- [Toward Effective Tool-Integrated Reasoning via Self-Evolved Preference Learning](https://mlsanity.com/paper/2509.23285/) (2025)
- [ET-Agent: Incentivizing Effective Tool-Integrated Reasoning Agent via Behavior Calibration](https://mlsanity.com/paper/2601.06860/) (2026)
- [Effective and Efficient Context Retrieval via Partial Dependency Graph for Repository-Level Code Generation](https://mlsanity.com/paper/2608.01927/) (2026)

## Abstract (arXiv)

This position paper proposes a fundamental shift in designing code generation models: treating reasoning depth as a controllable resource. Rather than being an incidental byproduct of prompting, we argue that the trade-off between rapid, direct answers ("fast thinking") and elaborate, chain-of-thought deliberation ("slow thinking") must be explicitly managed. We contend that optimizing reasoning budgets across the entire model lifecycle - from synthetic data creation and benchmarking to real-world deploymen - can unlock superior trade-offs among accuracy, latency, and cost. This paper outlines how adaptive control over reasoning can enrich supervision signals, motivate new multi-dimensional benchmarks, and inform cost-aware, security-conscious deployment policies. By viewing fast and slow thinking as complementary modes to be scheduled, we envision coding agents that think deep when necessary and act fast when possible.