Can you save on LLM tokens using images instead of text?
21 days ago
- #OpenAI API
- #LLM
- #Token Optimization
- Using images instead of text in LLM prompts can save tokens under specific conditions.
- Experiment showed a 40% reduction in prompt tokens with GPT-5 when using images.
- However, most models use more completion tokens with image inputs, offsetting savings.
- Completion tokens are more expensive, making the trade-off not worthwhile for most models.
- The ideal image size for OpenAI API is 768x768 to avoid resizing and maintain text readability.