Mistral AI announces that Le Chat, its AI-powered chatbot, is now available to everyone. While it used to take nearly 24GB of data to download in a torrent file to use Pixtral, the company has just bundled it into its web service.
Mistral AI's first multi-media AI in web version
Similar to ChatGPT or Google Gemini, Le Chat is available in a web version. Until now, it allowed dialogue with the Mistral Nemo, Codestral and Mistral Large 2 AI language models, and this new version gives free access to Pixtral 12-B, Mistral's first multimodal AI model.
In other words, a multimodal language includes its ability to process different data formats: for this language, it is the ability to analyze texts and images.
According to benchmarks published by Mistral AI, the startup is happy to match, and sometimes even outperform, some larger models, such as the LLaVA-OV 7B.
Pixtral Test: Generate HTML Code from a Diagram
We wanted to check out the capabilities of Pixtral 12-B highlighted by Mistral AI. The company claims that its language can generate computer code from a hand-drawn sketch. So we drew a web page on an iPad using the Procreate app using an Apple Pencil.
When we submitted this image to Pixtral, we included this prompt: “Write HTML code to create a site like this.” The chat is triggered and the source code is generated in HTML format.
It's strange that we were so quick to display the HTML code in the browser version.
Although the result may seem a bit superficial, the optical handwriting recognition is very effective. The layout is generally respected, except for the news sites which are not laid out as shown in the diagram.