Mixedbread has released Toast 1: a tool specifically designed to help large models gather data

According to Perceive Beating Monitor, the AI search company Mixedbread has released a dedicated search Agent Toast 1. It is mainly used to find information and filter sources for large models like GPT and Claude, and then return the results to the main model for processing.

Officially, Toast 1's search quality is said to be on par with cutting-edge closed-source models, but with faster speed and lower cost.

In Harvey's legal knowledge base test, after taking over the search, GPT-5.6 Sol's answer score remained the same, but the total cost was reduced by over 60%.

On Databricks OfficeQA Pro V2, GPT-5.6 Sol + Toast 1 scored 70%, higher than the previous 60% of Claude Fable 5 + Genie, at less than half the cost of the latter.

Toast 1 has been opened up through an API and can be integrated with other Agents specifically for search and information retrieval. The supporting search framework has been open-sourced, but the model weights have not been made available yet.