Previoulsy on GPT-4-vision was supported, change introduces support for Google/Anthropic and new OpenAI models Additionally this makes vision work properly in dev environments cause we sent the encoded payload via prompt vs sending urls