Accuracy is one consideration when choosing an LLM or improving retrieval-augmented generation (RAG). Larger models are generally more accurate, but can be slower and more expensive to run. section 1.5 section 8.1
Balance accuracy against practical constraints:
- Larger models generally offer more accuracy, while smaller models respond faster and cost less per token. section 1.5
- Choose according to the application: a customer-support chatbot may favor speed and cost, while legal document analysis may prioritize accuracy. section 1.5
- A workflow can use different models for different tasks to balance accuracy, speed, and cost. section 1.7
Choosing models across a workflow
A workflow might use GPT-5-nano for fast summarization, GPT-5-mini to synthesize an answer, and GPT-5 to route complex queries. section 1.7
For RAG, examine both content ingestion and question-answering: each stage can introduce problems or offer opportunities to improve accuracy. section 8.1
