Arabic input may use more tokens in Claude, according to a chart described by Abdul Osama (عبدول أسامة). He suggests trying an English version of a prompt and pasting text instead of sending a screenshot when the task only requires reading that text. These are options to test, not guarantees of lower usage for every task.
“You can translate it elsewhere, then give it the prompt in English.” — Abdul Osama
What is being compared?
Claude is an AI assistant made by Anthropic. Tokens are units into which text is divided for processing; the video focuses on how many are used when content is sent to the model.
| What you send to Claude | Observation in the video | What to try | | --- | --- | --- | | An Arabic prompt | It may use more tokens | Translate the prompt outside Claude and send it in English | | A screenshot containing text | Abdul noticed much higher token use | Copy text from the page or image and explain the task alongside it |
In numbers
- 2.74 times: the figure Abdul says the chart shows for tokens used to read Arabic in Claude. The chart is not provided here, so its measurement method and comparison conditions are unclear.
Background
Token counts can vary by language and by how content is submitted. Sending an image to be read is also different from sending text copied from it. Comparing actual usage on the same task is more useful than assuming one approach will always cost less.
What to do next
For a long prompt, try both the Arabic original and an English translation, then compare token usage and answer quality. If the information is text in a page or picture, paste that text with the necessary context when you do not need Claude to analyze the image’s visual elements.
Claude
