Guides 6 min read· 26 Jul 2026· By Guides
Vision inputs — sending images to GPT-5 and Gemini 3
URL, base64, batch — the three ways to hand a model a picture.
Vision is now table stakes. All chat models on TokenBaazar accept image_url inputs.
Via URL#
python
messages=[{"role": "user", "content": [
{"type": "text", "text": "What is in this photo?"},
{"type": "image_url", "image_url": {"url": "https://..."}}
]}]Via base64#
Prefix with data:image/png;base64,. Same field, inline data. Use for private uploads that shouldn't hit a public URL.
Images are billed by tile size — a 1024×1024 image ≈ 765 tokens on GPT-5.