French AI startup Mistral has recently implemented several updates to its product lineup to maintain its competitive edge in the highly competitive AI sector.
The Le Chat chatbot platform by Mistral now includes a web search feature that provides cited results, akin to OpenAI's ChatGPT. Additionally, Le Chat has introduced a "canvas" tool similar to ChatGPT Canvas, allowing users to modify, transform, or edit content using Mistral's AI models, such as web templates and data visualization charts.
According to a blog post by Mistral, users can utilize the canvas feature to create documents, presentations, code, web templates, and more. They can also edit content in place without needing to regenerate responses, manage draft versions, and preview designs.
In addition to the aforementioned features, Le Chat can now manage large PDF documents and images, performing analysis and summarization on files that include charts and formulas. The platform has integrated Flux Pro models from Black Forest Labs for image generation. Furthermore, Le Chat supports hosting shareable automated workflows designed for tasks like scanning expense reports and processing invoices, which Mistral refers to as "agents."
Some of Le Chat's new features will be available for free during the testing phase, enabled by Mistral's latest models.
Among them, Pixtral Large is a model capable of handling both text and images, and it is the second model in Mistral's Pixtral series. With 12.4 billion parameters, this model performs on par with or better than leading models in certain multimodal benchmark tests, including Anthropic's Claude 3.5 Sonnet, Google's Gemini 1.5 Pro, and OpenAI's GPT-4. The number of parameters generally correlates with a model's problem-solving capabilities, with more parameters typically indicating better performance.
Mistral has also launched a new version of its flagship text model series, Mistral Large 24.11. According to Mistral, the new version exhibits "significant improvements" in understanding lengthy content, making it suitable for document analysis and task automation scenarios.
Pixtral Large and Mistral Large 24.11 are available under two types of licenses: a restrictive license for research purposes and an enterprise license for development and commercialization. Mistral Large 24.11 is accessible via Mistral's API and the AI platform Hugging Face, and will soon be available through cloud platforms like Google Cloud and Microsoft Azure.
Based in Paris, Mistral has recently secured $640 million in venture capital and is progressively expanding its AI product offerings. In recent months, the company has introduced free model testing services for developers, an SDK that allows clients to fine-tune models, and new models including Codestral for code generation.
Mistral was founded by alumni from Meta and DeepMind with the aim of developing competitive models and associated services while achieving profitability in the process. Although achieving profitability is a challenge for all generative AI startups, it is reported that Mistral began generating revenue this summer.
