Skip to content
AI Atlas
PaperActive

SpecBench: Measuring Reward Hacking in Long-Horizon Coding Agents

arxiv.org/abs/2605.21384

Updated 50 min ago · first seen 11 Sept 2026

paper_01M294GQ30SS9Q0RRNWRE39CJ4

Published
11 Sept 2026
T1 · 50 min ago
arXiv
2605.21384
T1 · 50 min ago
Category
cs.SE
T1 · 50 min ago

Abstract

-cross Abstract: As long-horizon coding agents produce more code than any developer can review, oversight collapses onto a single surface: the automated test suite. Reward hacking naturally arises in this setup, as the agent optimizes for passing tests while deviating from the users true goal. We study this reward hacking phenomenon by decompose software engineering tasks into three parts: (i) a natural language description of the specification (ii) visible validation tests that exercise specified features in isolation, and (iii) held-out tests that compose those same features to simulate real-world usage. Based on the specification and the visible validation test suites, a genuine agent would be able to generate a solution that can also pass all of the held-out tests. Therefore we use the gap in pass rates on these two suites to quantify reward hacking. Based on this methodology, we introduce SpecBench, a benchmark comprising 30 systems-level programming tasks ranging from short horizon tasks like building a JSON parser to ultra long horizon tasks like building an entire OS kernel from scratch. Large-scale experiments reveal a consistent pattern: while every frontier agent saturates the visible suite, reward hacking persists, with smaller models exhibiting larger gaps on holdout suites. The gap also scales sharply with task length: it grows by 28 percentage points for every tenfold increase in code size. Failures range from subtle feature isolation to deliberate exploits, including a 2,900-line hash-table "compiler" that memorizes test inputs. SpecBench offers a principled testbed for measuring whether coding agents build genuine working systems or merely game the test suites developers hand them.

Authors 4

Bingchen Zhao, Dhruv Srikanth, Yuxiang Wu, Zhengyao Jiang

Specification

Official page

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Arxiv announce type
replace

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

arXiv id
2605.21384

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Categories
cs.SE, cs.AI, cs.CL

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

PDF

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Primary category
cs.SE

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Published
11 Sept 2026

Source:arXiv (Atom API + RSS)T1observed 50 min agohigh

Each value shows its source, tier and observation time. Conflicting claims are kept side by side and flagged — never averaged. How AI Atlas records facts →

Provenance

Attributed facts

9

Source tiers

T19

Freshest observation

50 min ago

Conflicts

None