Vicuna is an open-source chat robot trained by fine-tuning LLaMA on user-shared conversations. Initial evaluations using GPT-4 as a judge showed that Vicuna-13B achieved 90% of the quality of OpenAI ChatGPT and Google Bard in over 90% of cases, and outperformed other models like LLaMA and Stanford Alpaca in over 90% of cases. The training cost of Vicuna-13B is approximately $300. The code, model weights, and online demo are available for non-commercial use.