Baichuan Intelligence announced two API based on search enhancement at the Baichuan2 Turbo series API launch event: Baichuan2-Turbo-192K and Baichuan2-Turbo. These two APIs not only support a long context window of 192K, but also utilize the functionality of the search enhancement knowledge base.

Baichuan Intelligence's CEO, Wang Xiaochuan, stated that this method allows large models to find the most relevant content to user prompts from a massive amount of documents, and then put them into a long window together with the prompts, thereby improving the scale of text and inference efficiency. He said that this can increase the text scale from 192K to the equivalent of 50 million tokens, which is equivalent to 100 million Chinese characters.

Baichuan Intelligence also announced that API users can upload their own text materials to create exclusive knowledge bases. The official website's large model already supports various text formats and URL inputs. Users can experience the improved model through the official website.