Gemini 4 Carbon is now the top overall vision model on our benchmarks
Text/OCR seems to be a consistent high performing area over others
- text localization and recognition: finding text, drawing boxes, and reading it
- short string extraction: reading a specific value from an image
- long text transcription: reading full passages from images
Beyond that, it's #1 in core vision tasks like object detection and reasoning.
Disclaimer: it is the slowest and most expensive of the top models so if you need cost/speed, this one isn't for you
Text/OCR seems to be a consistent high performing area over others
- text localization and recognition: finding text, drawing boxes, and reading it
- short string extraction: reading a specific value from an image
- long text transcription: reading full passages from images
Beyond that, it's #1 in core vision tasks like object detection and reasoning.
Disclaimer: it is the slowest and most expensive of the top models so if you need cost/speed, this one isn't for you
0:06
Gemini 4 Carbon is so fucking good at computer visionView quoted post on X
- #1 overall
- #2 object detection
- #1 short string extraction
- #4 structured JSON extraction
- #1 long text transcription
- #1 text localization and recognition
very slow and expensive though...
↓ examples