{
  "id": 15297201,
  "title": "I run OpenCode with a different model for every job, and I haven't hit a rate limit yet",
  "url": "https://urgent.news/2026/10/17/i-run-opencode-with-a-different-model-for-every-job-and-i-havent-hit",
  "topic": "tech",
  "section": "Tech",
  "published": "2026-10-17T23:00:20.000Z",
  "source": {
    "name": "XDA Developers",
    "slug": "xda-developers",
    "url": "https://www.xda-developers.com/run-opencode-with-different-model-for-every-job-havent-hit-rate-limits/"
  },
  "original_language": "en",
  "account": "If you utilize Claude daily or for high-demand projects, you've likely monitored the usage meter as closely as the actual output. Anthropic has increased the limits several times this year, yet a weekly cap remains in place for all models, with Claude Code and the chat app pulling from the same pool. The current model I favor, Opus 5.5, depletes this weekly allowance quicker than lighter alternatives. I've been treating OpenCode as a more affordable backup for when Claude is out of commission, but using a free model when a paid one runs out still consumes one usage bucket at a time. To address this, I decided to be more intentional about assigning tasks to different models instead of picking one out of habit. Claude has two limit types - a rolling five-hour session window and a weekly cap that covers every model under Pro. While the five-hour side has improved over the year, doubling Claude Code's limits in May and again with Opus 5.5, the weekly cap hasn't increased, causing the weekly allowance to drain faster. Although Opus 5.5 is my preferred model, I've been less discerning with it lately, even using it for basic tasks like spell-checking. Shifting to a lighter Claude model does help alleviate some of the strain on the weekly cap, but the issue persists across all accounts. I use ChatGPT on OpenCode as well, which connects seamlessly to my OpenAI subscription, allowing me to select GPT-6.1 Sol for projects that require building from scratch. The free models, such as Nemotron 3.5 Super VL, Big Pickle, and GLM-5.3-Flash, handle smaller edits and questions, while I strategically assign complex tasks to the premium models. This approach ensures that the premium model focuses on high-value projects, while the free models handle lighter workloads. By utilizing different models for various tasks, I find the process more efficient and honest, as free models often provide less predictable results. For instance, Big Pickle struggled with a minor edit when left to its own devices, but when I handed the same prompt to Nemotron, the task was completed in under a minute. Claude's speed issues have also made me less reliant on it for tasks that require quick turnaround. Additionally, privacy concerns have influenced my model selection. Free model data may be used to improve the model during its free period, and NVIDIA's Nemotron endpoints are trial-only, so I avoid sending sensitive information there. Furthermore, Claude isn't an option within OpenCode's subscription, which I would change if possible. While Opus remains my top choice, my usage of Claude has decreased, and I now allocate tasks across different models to maximize value and minimize rate limits.",
  "summary": "I stopped leaning on one model for everything",
  "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."
}