Mistral AI is preparing for a new model release, which potentially would have image generation and web search capabilities. Some of these features have been spotted being in development earlier this month. However, now there are also traces of a new model called "Pandragon", with image generation and web search features that seem to be associated with this exact model in particular.
Both of these features are also marked as Beta, so it’s very possible that the release or rollout might be limited to a certain group of users or it will be just a label telling the users that it’s still in the experimental stage. According to previous discoveries, there is a possibility that the search will be powered by Brave index as well as AFP fact-checking platform, whereas image generation reportedly may be powered by Flux from BlackForestLabs.
There are not yet many details about this model, but it seems like it might be the same Mistral Large 2 trained to work with these new tools. Besides these, there are also a couple of new things that can be found in the code. One of them is called connectors, which potentially may signal the integration with Slack, which was discovered earlier as well.
Even other similar products like ChatGPT typically would also include integration with Google Drive or OneDrive as well for file upload. This also correlates with the new vision capability, which will allow users to upload documents in the future as well. It’s also possible that this document upload feature will be released separately because it has a separate feature flag.
As a reminder, currently, it’s only possible to upload images to Mistral and only Pixtral model has this capability. However, it seems that for documents, both Pixtral and Mistral Large 2 may get this feature. In general, this is a very exciting addition. It’s still unclear when and in which order these features will be released.
