AI Prompt Library
Categorized production-ready prompts mapped to the most suitable models, with token estimates and cost per run. Copy any prompt, paste it into your API call or chat interface, and see the estimated cost before you scale.
Filter Prompts
Prompts
Prompt Engineering Best Practices
- Be specific: "Summarize in 3 bullet points under 20 words each" beats "Summarize this".
- Use structured output: Ask for JSON when you need to parse the response programmatically.
- Provide context: Include relevant background, constraints, and examples in the system prompt.
- Cache the system prompt: Move static instructions to a cached prefix to save 50-90% on repeat calls.
- Limit few-shot examples: 2 examples is usually enough; more rarely improves quality proportionally.
- Set max_tokens: Cap output to the minimum you need — output tokens are 4-8× more expensive than input.
- Use the right model: Don't use a flagship for classification. Use a nano model for routing and a flagship for complex reasoning.
FAQ
Are these prompts free to use?
Yes. All prompts in this library are free to copy and use in your own projects, API calls, or chat interfaces. No attribution required.
How is the cost per run calculated?
Cost = (estimated input tokens × input price) + (estimated output tokens × output price) for the selected model. Switch models to compare costs. Actual costs vary based on your real input length and output length.
Which model should I use for each prompt?
Each prompt lists recommended models. As a rule: use economy models (GPT-5 Nano, DeepSeek V3) for simple tasks, balanced models (GPT-4.1 Mini, Claude Sonnet 4) for most work, and flagships (GPT-5, Claude Opus 4) for complex reasoning.