{
  "id": 21069812,
  "title": "Where an agent's tokens go: 9 real GitHub bugs, two Gemini Flash-Lite models, 1 fix each",
  "url": "https://urgent.news/2026/11/21/where-an-agents-tokens-go-9-real-github-bugs-two-gemini-flash-lite",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-21T16:11:01.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/linqiangmax/where-an-agents-tokens-go-9-real-github-bugs-two-gemini-flash-lite-models-1-fix-each-32j1"
  },
  "original_language": "en",
  "account": "Two Gemini Flash-Lite models were given nine real GitHub bug reports to fix, with each model fixing one bug. Between them, they made 964 model calls during the process. The tasks were sourced from SWE-bench-Live, using small pure-Python repositories that could be run on a Windows machine without Docker. The model calls were made using the free tier versions of Gemini 3.5 and 3.1, resulting in different costs based on token usage. Overall, the agents spent a total of approximately $0.56 to $0.43 per task to fix the bugs, with a cost of about 43 to 56 cents per fixed bug.",
  "summary": "I gave two Gemini Flash-Lite models the same 9 real bug reports from GitHub and let each one work in a loop: search, read a file, edit, run the tests, submit. Each model fixed one bug. Between them they made 964 model calls. The tasks come from SWE-bench-Live (github.com/microsoft/SWE-bench-Live, MIT). I picked small pure-Python repos I could run on a Windows machine without Docker: attrs, faker…",
  "key_points": [
    "Two Gemini Flash-Lite models fixed nine real GitHub bugs each",
    "Agents made 964 model calls during the bug-fixing process",
    "Agents spent $0.56 to $0.43 per task to fix the bugs"
  ],
  "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."
}