Instruction file imported from theeufj/RNSR (.cursor/rules/llm-model-validation.mdc). Copyright stays with the author.

LLM Model Validation

Before setting up LLM calls, recommending models, or modifying LLM configuration code, you MUST verify that the model name is current and valid.

Required Verification Steps

1. Check official documentation using web search before recommending any model:

   • OpenAI: Search for current model names at docs.openai.com
   • Anthropic: Search for current model names at docs.anthropic.com
   • Google Gemini: Search for current model names at ai.google.dev

2. Never assume model names - Model names change frequently:

   • Models get deprecated (e.g., gpt-4-turbo -> gpt-4o)
   • Version formats change (e.g., gemini-pro -> gemini-1.5-pro -> gemini-2.0-flash)
   • Preview models expire (e.g., gemini-3-flash-preview may not exist)

3. Verify fallback models too - When implementing fallback logic, verify all model names in the fallback chain

Example Verification

Before writing code like:

DEFAULT_MODELS = {
    "openai": "gpt-4o-mini",
    "gemini": "gemini-2.5-flash",
}

First search: "OpenAI current model names 2026" and "Gemini current model names 2026"

Common Mistakes to Avoid

• Using gemini-2.5-flash when it may be deprecated
• Using preview model names that have expired
• Assuming model naming conventions without checking
• Copying model names from old code examples

When in Doubt

Ask the user: "Which specific model would you like to use? I can verify it exists in the current API."