{
  "id": 19844695,
  "title": "DeepSeek’s coding agent is too tidy for its own good — and benchmarks are punishing it",
  "url": "https://urgent.news/2026/11/13/deepseeks-coding-agent-is-too-tidy-for-its-own-good-and-benchmarks",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-13T17:56:30.000Z",
  "source": {
    "name": "The New Stack",
    "slug": "the-new-stack",
    "url": "https://thenewstack.io/deepseek-harness-cleanup-benchmark/"
  },
  "original_language": "en",
  "account": "DeepSeek's open-source coding agent, DeepSeek Harness, is underperforming on a major benchmark due to its habit of shutting down background processes when a task finishes. This behavior is causing it to drop from first to sixth place on Terminal-Bench, a benchmark used to compare coding agents. DeepSeek Harness is statistically tied for first across the board's five scored evaluations, but it only passed two out of 36 attempts using DeepSeek-V4.1-Flash and GLM-5.3-Flash, while other agents like OpenCode, Pi, and Hermes Agent passed all 36. This behavior is costing DeepSeek Harness between 1.5 and 1.7 percentage points on the benchmark. DeepSeek Harness' documentation states that the local job registry is intended as a short-lived component, designed to die with the harness process. The behavior is a deliberate design choice, meant for jobs that should live in the harness process and die with it. Other harnesses in the leaderboard show substantial differences in performance when the same model and reasoning effort are used. DeepSeek Harness is cheaper than some alternatives like Pi, but less cost-effective than others like OpenCode. The New Stack contacted DeepSeek for comment, but had not received a response by the time of publication.",
  "summary": "DeepSeek’s open-source coding agent is losing points on a major benchmark for a habit developers usually treat as good practice: The post DeepSeek’s coding agent is too tidy for its own good — and benchmarks are punishing it appeared first on The New Stack .",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}