C.W.K.
Stream
Lesson 02 of 05 · published

Image Detail Levels

~22 min · vision, detail, cost

Level 0Tokenizer
0 XP0/54 lessons0/10 achievements
0/120 XP to next level120 XP to go0% complete

The detail parameter controls image resolution and directly affects token cost.

LevelDescriptionToken Cost (gpt-4o)
lowResized to 512×512, fixed cost~85 tokens
highFull resolution, tiled analysis~765 tokens (1024×1024)
autoModel decides based on contentVaries

Token Cost Calculation (high detail)

For gpt-4o, gpt-4.1, gpt-5 (tile-based):

  1. Scale image to fit within 2048×2048
  2. Resize shortest side to 768px
  3. Count 512px tiles: tiles × 170 tokens + 85 base tokens
Model FamilyBase TokensPer Tile
gpt-5, gpt-5-chat70140
gpt-4o, gpt-4.185170
o1, o375150

Example: A 1024×1024 image with high detail on gpt-4o = 4 tiles × 170 + 85 = 765 tokens. The same image with low = 85 tokens. Use low when fine detail isn't needed.

External links

Exercise

Pick one document scan. Run OCR-style extraction at low, high, and auto. Record: latency, image_tokens, accuracy (manual). Pick a default for your app.

Progress

Progress is local-only — sign in to sync across devices.
Spotted a bug or have feedback on this page?Report an Issue

Comments 0

🔔 Reply notifications (sign in)
Sign inPlease sign in to comment.

No comments yet — be the first.