The cost of processing image inputs for large language models varies significantly between providers, with different formulas for converting pixels into billable tokens. OpenAI's GPT-5.6, for example, charges per 32x32 pixel patch, while Anthropic uses 28x28 patches, and Google's Gemini 3 has a flat rate based on media resolution. When tested with a typical phone photo, GPT-5.6, Claude Sonnet 5.5, and Gemini 3.1 Pro showed substantial cost differences, with Gemini being the most expensive at $2.24 per thousand images compared to OpenAI's $28.73. The article suggests that users can control costs by resizing images themselves before sending them to the model, as models can often read information from smaller, lower-resolution images. AI
IMPACT Understanding tokenization costs for image inputs can help developers optimize LLM usage and reduce operational expenses.
RANK_REASON The item analyzes and compares existing LLM features and pricing, rather than announcing a new release or significant industry event.
- Anthropic
- Claude-4.7
- Claude Sonnet 5.5
- Gemini
- Gemini 2.5
- Gemini-3.*
- Gemini 3.1 Pro
- GPT-4o
- GPT-5.6
- OpenAI
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
