Continuous maintenance improves reliability of coding benchmarks for ai models
Benchmarks as Software: Terminal-Bench and the Case for Continuous Benchmarks
Artificial Intelligence
Summary
Benchmarks measure how well AI coding agents perform, but over time, these tests can become outdated or flawed as models get better. The authors show that frequent checking and fixing of these tests, including adjusting resource limits, helps keep benchmark results accurate and meaningful. They applied this to a coding benchmark called Terminal-Bench, fixing many issues and showing how performance rankings can change as a result. Their work highlights that benchmark upkeep involves more than bug fixes—it also means rechecking assumptions and timing rules.
What this means in practice
- •For software evaluation teams: Maintain coding benchmarks by combining automated checks and human feedback to keep evaluation results reliable as AI models improve.
- •For cloud infrastructure managers: Adjust AI benchmark resource limits such as runtime to better reflect true model capabilities during evaluations.
- •For automated testing platforms: Create products that continuously update AI coding benchmarks to ensure accurate model performance tracking over time.$Commercial implications: Enables selling benchmark maintenance tools that integrate human and automated validation for AI coding model assessments.
Authors
E. Kelly Buchanan, Crystal Zhou, Scott W. Linderman, Christopher Ré
Abstract
Coding agents are among the most economically consequential deployments of LLMs to date, making reliable evaluation increasingly important. Yet even benchmarks that are carefully validated at release can become stale. As models improve, they can discover valid solutions that existing verifiers reject, while changes to evaluation infrastructure can invalidate assumptions embedded in tasks and tests. We call the resulting accumulation of defects benchmark maintenance debt: a growing gap between benchmark scores and the capabilities they are intended to measure. We introduce a continuous-validation workflow for agentic benchmarks that combines community reports, task artifacts, agent traces, automated oracle checks, and human-gated repairs. We apply this workflow to Terminal-Bench 2.0, a benchmark that underwent extensive human and LM-assisted review and red-teaming before release. Our audit identifies issues in roughly 30% of the benchmark, and we release the resulting repairs as Terminal-Bench 2.1. These repairs change evaluation results: accuracy increased by over 10 percentage points and leaderboard positions move by up to three ranks, even as the overall ranking remains highly correlated (Spearman rho = 0.958). We evaluate how resource allocation shapes benchmark conclusions. We show that fixed wall-clock can materially change agent performance: For example, extending the time limit closes a 20pp harness gap on GPT 5.4 and shrinks a 30 pp gap on GPT 5.4 mini to 5 pp. Benchmark maintenance is therefore not only about repairing task artifacts: resource budgets are part of the measurement specification and must be validated alongside the tasks themselves. More broadly, benchmark maintenance debt is not just accumulated implementation bugs: our results show that as models improve, both evaluation assumptions and resource settings may need to be revalidated.