The French startup Mistral AI has made a significant leap in the world of artificial intelligence with the launch of Pixtral Large, a multimodal model with 124 billion parameters. This system positions itself as one of the most advanced on the market, offering leading capabilities in understanding texts, images, graphs, and complex documents, surpassing models like GPT-4o and Gemini 1.5 Pro in key tests.
Pixtral Large: Innovation in Multimodal Understanding
Pixtral Large stands out for its ability to simultaneously process up to 30 high-resolution images or a 300-page book, thanks to its expanded context window of 128,000 tokens. This model, based on the successful Mistral Large 2, combines a vision encoder with 1 billion parameters and a multimodal decoder with 123 billion parameters, maintaining its leadership in textual tasks while expanding its domain to complex visual data.
Among the model’s notable achievements are:
- MathVista: 69.4% accuracy in visual mathematical reasoning, outperforming all current models.
- DocVQA and ChartQA: Better performance in document and graph understanding compared to GPT-4o and Gemini 1.5 Pro.
- MM-MT-Bench: Leader in this evaluation designed to measure performance in real-world multimodal use cases.
Le Chat: A Comprehensive Workspace Powered by Pixtral Large
Mistral’s Le Chat platform has also received significant updates, transforming into a comprehensive environment for content creation and management. Highlights include:
- Integrated web search.
- Advanced document analysis.
- Image generation, thanks to the Flux Pro technology from Black Forest Labs.
- Canvas: A real-time content creation and editing tool, similar to those offered by OpenAI and Anthropic.
During its beta phase, these features will be available for free, allowing users to experience cutting-edge capabilities without financial barriers.
Availability and Licensing
Pixtral Large is available under two types of licenses:
- Mistral Research License (MRL): for academic and educational use.
- Commercial License: for testing, development, and commercial applications.
The model can be tested directly on the Le Chat platform, downloaded from Mistral’s official site, or integrated through its APIAn API, short for “Application Programming Interface,”… as pixtral-large-latest.
Impact on the Global AI Landscape
With this launch, Mistral positions itself as a strong competitor in a market historically dominated by American companies. By offering open-source and accessible models, the French company underscores its commitment to more inclusive and collaborative AI, marking a shift in the competitive dynamics of the sector.
Pixtral Large not only reinforces Mistral’s leadership in developing multimodal models but also demonstrates how European innovation can challenge tech giants, offering advanced solutions for real use cases in sectors like finance, healthcare, research, and more.
Mistral Large 24.11: Continuous Improvements in Textual Understanding
In addition to Pixtral Large, Mistral has launched an update of its flagship textual model, Mistral Large 24.11, now available on platforms such as Google Cloud and Microsoft Azure. This model improves the understanding of long contexts, introduces an optimized prompt system, and provides more precise features, making it ideal for business workflows like task automation and knowledge exploration.
With Pixtral Large and Mistral Large 24.11, Mistral AI reaffirms its commitment to innovation in artificial intelligence, offering powerful and accessible tools that promise to transform the way we interact with complex data in multiple formats.
Source: Artificial Intelligence News