{
  "id": 1286285,
  "title": "The Official DeepSWE Number Is Footnoted to DeepSeek Harness",
  "url": "https://urgent.news/2026/08/16/the-official-deepswe-number-is-footnoted-to-deepseek-harness",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-16T14:39:47.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/reidmarlow/the-official-deepswe-number-is-footnoted-to-deepseek-harness-1chj"
  },
  "original_language": "en",
  "account": "On August 13, DeepSeek released four significant updates in a single day. First, the V4-Pro model preview was made available. Next, peak and off-peak API pricing began at 16:00 UTC on August 16. A 88-page draft paper on spatiotemporal composability was also published. Finally, deepseek-harness was released as a developer preview under the MIT license, with a single command to run: npx @deepseek-ai/dsh web.\n\nThe official DeepSWE number for the DeepSeek-V4-Pro-0813 model was 62.7, compared to 12.8 for the preview version. This figure is footnoted to indicate that the testing of public code-agent sets was conducted within the DeepSeek Harness in minimal mode.\n\nAdditionally, Fable 5 maintained its score of 70.0 on the same grid. A loop plugin was introduced, and its implementation was named Cordis. The peak output for V4-Pro was reported at $3.96 per million tokens.\n\nTo illustrate these points, a four-minute video was created, covering the footnote details, the plugin topic, and the peak-hour billing information. The original August 13 write-up focused on the plumbing version, which framed DeepSeek's Harness as the price signal. This story presents the scoreboard version, emphasizing the key metrics and updates.\n\nFor those already using Claude Code with a DeepSeek key, the article raises the question of whether the plugin topic alters their experience.",
  "summary": "On August 13, DeepSeek shipped four things in one day. V4-Pro left preview. Peak and off-peak API prices start at 16:00 UTC on August 16. An 88-page draft paper went up on spatiotemporal composability. And deepseek-harness dropped as a developer preview, MIT license, one command: npx @deepseek-ai/dsh web . The model headlines ate the English YouTube tab. I stayed on the harness. Official figure…",
  "key_points": [
    "DeepSeek released four updates on August 13",
    "DeepSWE number 62.7 for DeepSeek-V4-Pro-0813 model",
    "MIT-licensed deepseek-harness available for preview"
  ],
  "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."
}