The French startup Mistral AI has launched a new model called "Pixtral Large," a multimodal AI with 124 billion parameters, positioning itself as a serious competitor to OpenAI.
Pixtral Large can process both text and visual data effectively, making it suitable for interpreting charts and analyzing documents. It features a context window of 128,000 tokens, allowing it to handle up to 30 high-resolution images or a 300-page book.
Arthur Mensch, CEO of Mistral AI, stated on platform X, "At Mistral, we realized that models and product interfaces must be designed to create the best AI experience. Pixtral was trained with high-performance front-end applications in mind."
The model is available for free download on Hugging Face under a special Mistral AI Research License, limiting its use to research applications. Commercial use requires an API through Mistral's managed web service or a separate license.
Mistral AI has also updated its free AI tool, "Le Chat," incorporating new features from Pixtral Large. This tool can now perform real-time web searches, analyze complex PDFs, and automate tasks through customizable AI agents.
Despite its advancements, Le Chat lacks advanced audio capabilities found in competitors like OpenAI's ChatGPT Advanced Voice Mode. Mistral aims to differentiate itself with a comprehensive feature set while remaining free, unlike many similar premium services.
As Mistral AI grows, it may establish itself as a leading European AI player, especially in light of shifting dynamics in the tech world.