Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.04433

What Does a Harness Buy? Tokens, Mostly

Yangze Liu·Zhongyi Han
View original source ↗Read source document ↗Export metadata ↓

Abstract

A coding agent is a language model wrapped in a harness: the system prompt, the tool set, and the context management that turn a chat model into something that can work inside a repository. Production harnesses ship releases daily, vendors advertise pass-rate gains from harness changes, and leaderboards mix harnesses freely. What is rarely measured is how much the harness itself moves the score when the model is held fixed. We run five models through three production harnesses, Claude Code, mini-SWE-agent, and OpenCode, on SWE-bench Verified, and rerun the same configurations to calibrate how much a score moves when nothing changes but the run. On 447 tasks and the two models we ran there, Claude Code and mini-SWE-agent, the heaviest and the lightest harness, are equivalent within five points. On a 45-task hard subset and five models, swapping the harness flips as many tasks as rerunning the same harness, 13% in both cases, and the tasks a harness wins in one run are not the tasks it wins in the next. The one harness effect that clears the noise is a loss, not a gain: OpenCode trails by up to 9 points on the large pool, and on one model half of that gap sits in runs its output cap cut short. What the harness does decide is the bill. With the same model, the same tasks, and one price list, cost per task differs by up to 3x across harnesses. The gap is set at the first call, by the preamble of system prompt and tool schemas each harness sends with every step, and scaled by the number of steps; per-step growth and per-call tool output differ far less. The provider's price for cached input scales the bill and does not reorder it. The rerun data also give the resolution a harness comparison needs: at the discordance we observe, 45 tasks catch a 13-point gap only half the time and no gap with 80% power, and 447 tasks resolve 5 points, still coarser than the gains many harness changes claim.

Explore related subjects

cs.AIcs.LGcs.SE

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Yangze Liu, Zhongyi Han. 2026-10-03. What Does a Harness Buy? Tokens, Mostly. https://arxiv.org/abs/2610.04433

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

When Explanations Compete: Policy-Aware Selection Under Uncertainty

Uncertainty-aware explanation methods often produce several alternatives for the same prediction. Selecting among them requires a policy for balancing prediction confidence, uncertainty, and application constraints. This paper presents a framework for applying such policies to a fixed set of generated explanations. Candidates are characterised by uncertainty change, prediction direction, and, when available, interval position relative to a decision boundary. The framework combines these properties with eligibility rules, optional bidirectional Pareto screening, and policy-aware ranking. A fictitious prostate-cancer example illustrates how different explanatory purposes lead to different selections from the same candidate set. We instantiate the framework with Calibrated Explanations for classification, thresholded regression, and plain regression. Across 41 benchmark datasets, mean candidate counts range from 11.57 to 21.75 for single-feature explanations and from $29.48$ to $69.53$ when conjunctions are included. Equal-weight and confidence-only policies yield an average selection-disagreement rate of $28.7\%$ while favouring the same confidence direction. A supporting $δ$-CLUE experiment demonstrates use with a second generator. By making the selection policy explicit, the framework allows applications to compare and prioritise explanations according to their intended use.

cs.AI↗
arXiv2026

DrugMCTS: a drug repurposing framework combining multi-agent, RAG and Monte Carlo Tree Search

Recent advances in large language models have demonstrated considerable potential in scientific domains such as drug repositioning. However, their effectiveness remains constrained when reasoning extends beyond the knowledge acquired during pretraining. Conventional approaches, such as fine-tuning or retrieval-augmented generation, face limitations in either imposing high computational overhead or failing to fully exploit structured scientific data. To overcome these challenges, we propose DrugMCTS, a novel framework that synergistically integrates RAG, multi-agent collaboration, and Monte Carlo Tree Search for drug repositioning. The framework employs five specialized agents tasked with retrieving and analyzing molecular and protein information, thereby enabling structured and iterative reasoning. Extensive experiments on the DrugBank and KIBA datasets demonstrate that DrugMCTS achieves substantially higher recall and robustness compared to both general-purpose LLMs and deep learning baselines. Our results highlight the importance of structured reasoning, agent-based collaboration, and feedback-driven search mechanisms in advancing LLM applications for drug repositioning.

cs.AI↗
arXiv2026

PuzzleJAX: A Benchmark for Reasoning and Learning

We introduce PuzzleJAX, a GPU-accelerated puzzle game engine and description language designed to support rapid benchmarking of tree search, reinforcement learning, and LLM reasoning abilities. Unlike existing GPU-accelerated learning environments that provide hard-coded implementations of fixed sets of games, PuzzleJAX allows dynamic compilation of any game expressible in its domain-specific language (DSL). This DSL follows PuzzleScript, which is a popular and accessible online game engine for designing puzzle games. In this paper, we validate in PuzzleJAX several hundred of the thousands of games designed in PuzzleScript by both professional designers and casual creators since its release in 2013, thereby demonstrating PuzzleJAX's coverage of an expansive, expressive, and human-relevant space of tasks. By analyzing the performance of search, learning, and language models on these games, we show that PuzzleJAX can naturally express tasks that are both simple and intuitive to understand, yet often deeply challenging to master, requiring a combination of control, planning, and high-level insight.

cs.AI↗