Hopefully not a stupid question... I notice most agents are using Sonnet

[nategrech]

With Opus 4.1 available, would certain agents benefit from using Opus? The .md references "sonnet" but not specifically which model? Should I be nominating which model specifically? Do you think Opus is suitable for any of the more demanding tasks? Newbie here so be kind :) Thank you in advance.

[lst97]

Hi @nategrech, thank you for your question, I am trying to explain my though with my limited knowledge.

TLDR: opus for planning, sonnet for execution

---

1. For most agents, I find that Sonnet provides sufficient performance. While you can upgrade to Opus for a boost in quality, be aware that it comes at a significant trade-off, consuming approximately 3 to 5 times [1] more tokens and increasing processing time.
   Therefore, using Opus is most beneficial for agents that handle complex reasoning, such as those designed for debugging, architectural design, and planning. For routine development agents like react-pro or python-pro, Sonnet remains the more efficient choice. Ultimately, there is no single best answer; you must carefully balance your desired output quality against your token budget when deciding which model to use. Of course, if the budget allowed, turning all agents to opus will definitely improve overall quality.

2. According to the documentation, model config does not require providing version number, therefore I assume it will be using the latest Claude model by default.

I will keep sonnet as the default model in this repo, but you can change it accordingly to balance between quality and token usage.

Here is more detail explain from AI (Gemini 2.5 Pro) for your reference to what model you should use base on situations:

When to Use the Advanced Model (e.g., Claude Opus 4.1)

Use the most powerful model for tasks that require creativity, complex problem-solving, and a deep understanding of context. Think of it as your "senior architect" or "tech lead."

• Complex Architecture and System Design: When you need to design a system from the ground up.

  Example: "Design a scalable and accessible component library architecture for a multi-brand design system using React and Storybook."

• Debugging Obscure or Ambiguous Bugs: For errors that aren't easily solved by a quick search, especially those involving state management, race conditions, or complex asynchronous logic.

  Example: "My React application has a memory leak related to useEffect cleanup in a component using WebSockets. Here is the code; find and fix the leak."

• High-Level Prototyping & Greenfield Projects: When you have a vague idea and need a "thought partner" to explore different implementations and generate foundational code.

  Example: "I need a proof-of-concept for a real-time collaborative whiteboard using Next.js, Zustand for state, and PartyKit for networking. Generate the initial project structure and key components."

• Large-Scale, Context-Aware Refactoring: When you need to refactor a significant part of a codebase while understanding the interconnected dependencies.

  Example: "Refactor our entire legacy Redux implementation to use React Query and Zustand, ensuring minimal disruption to existing features."

When to Use the Specialized/Balanced Model (e.g., Claude Sonnet 4)

Use the faster, more affordable model for everyday, high-volume tasks that are well-defined and repetitive. Think of this as your "pair programmer" that handles the grunt work efficiently.

• Generating Boilerplate and Scaffolding: Creating standard files, components, or tests.

  Example: "Create a responsive Card component in React and TypeScript with props for title, imageUrl, and description. Also, generate a corresponding Storybook story and a basic Jest/RTL test file."

• Writing Utility Functions and Snippets: Creating isolated, single-purpose functions.

  Example: "Write a TypeScript utility function formatCurrency(value: number, currency: string) that formats a number into a localized currency string."

• Simple, Contained Refactoring: Making straightforward code improvements or translations.

  Example: "Convert this React class component into a functional component using hooks," or "Translate this CSS module into Tailwind CSS classes."

• Documentation and Comments: Generating explanations for existing code.

  Example: "Add JSDoc comments to this complex function, explaining its parameters, return value, and potential side effects."