fix: set maxTokens to 4096 so vision model is not stopping too early (seems to...

fix: set maxTokens to 4096 so vision model is not stopping too early (seems to have a lower default than other models)