An image is billed as tokens
On Groq an image counts as 2,048 input tokens. Against the free limit of 8,000 tokens per minute, that is three images a minute with almost no room for text. The request limit is rarely what stops a vision workload on a free tier; the token limit is.
Vision is a property of the model, not of the provider
Groq’s free table has one model with image input, qwen/qwen3.8-27b, limited to three images per request. On OpenRouter the models API says, for each model, which inputs it accepts; the table above is filtered from it and carries its collection time.
Think twice about what is in the picture
Screenshots and photographed documents often contain personal data. On the Gemini API free tier, content is used to improve Google products and may be read by human reviewers. See data and training before sending real documents.