Weighing DeepSeek V4 against GPT-5.4 mini for lower-cost RAG

The planned RAG system would search a public of articles for each user question. It would score and rank the matches, then use the highest-ranked articles to create an answer. GPT-5.4 mini is one option, but its price could become a problem as usage grows.

is being considered because its price and appear attractive. No test results or firm conclusions are provided about answer quality, , or how well either model uses retrieved .

Key points

  • The system would search a public article for each question.
  • It would score and rank results before using the best matches to create an answer.
  • GPT-5.4 mini may become costly when usage increases.
  • is being considered for its price and .
  • No direct quality or total-cost comparison is included.
Read original