Mistral AI to Launch Multimodal Large 2.1 Model Using Canvas and Other Tools

Mistral AI has recently enhanced its Le Chat tool by implementing significant updates and outlining future development plans. The former model selector feature has been eliminated in favor of a more intelligent automatic selection mechanism. In this revamped setup, typical inquiries will be handled by Mistral 2 by default, whereas those containing image attachments will automatically be directed to Pixtral for processing. Furthermore, users retain the ability to designate custom proxies using the @mention shortcut, allowing for improved control over their chosen assistants.

Market rumors have emerged about a new multimodal model called Pandragon. Some analysts suggest that it might be associated with the forthcoming Mistral 2.1 release. It is anticipated that Pandragon (potentially Mistral 2.1) will offer multimodal functionalities, including tools for web searching and image generation among others. While it's not yet confirmed if Pandragon and Mistral 2.1 refer to the same model, hints within the code imply that Mistral 2.1 will incorporate support for multimodal features.

The currently released version of Le Chat features a concealed beta user interface that unveils three new functionalities, which are expected to become available to all users soon, potentially bypassing a private testing phase. Although these features are not fully implemented due to the absence of image preview capabilities, they are highly anticipated: 1. A web search function powered by Brave Search Index, distinguishing Mistral's search indexing from traditional engines like Bing or Google. 2. An image generation feature supported by Flux, operating similarly to Grok. 3. A collaborative interface named Canvas, enabling users to edit and refine generative content with preview and export options, akin to Anthropic's Artifacts or OpenAI's Canvas.

Additionally, two more features are anticipated to launch in the near future. One is PDF attachment support, which has been indicated in the update of the attachment button user interface, although it is not yet available. The other planned feature is a routing function called Connectors, which includes Notion Alpha integration. This could enable users to connect to Notion via API keys or allow Le Chat to retrieve data from Notion using Retrieval-Augmented Generation (RAG) technology. However, it is noteworthy that Notion Alpha does not appear in the beta feature list, suggesting it may be reserved for enterprise or team accounts.

Overall, Mistral AI continues to advance the feature set of Le Chat, enhancing its performance and differentiating itself through unique choices, such as utilizing Brave Search Index. While it remains competitive alongside major rivals, Mistral maintains a significant edge by open-sourcing its models and providing robust performance metrics and transparency in its development methods. This approach not only boosts user trust but also garners increased attention for Mistral in the highly competitive market.