French generative artificial intelligence startup Mistral AI has boldly stepped up to challenge industry giants like OpenAI with a series of groundbreaking announcements revealed today. Among the highlights is the introduction of new chatbot functionalities that significantly enhance its capabilities, surpassing those of ChatGPT. The company also unveiled a pair of formidable large language models (LLMs), including the debut of Pixtral Large and an upgraded version of Mistral Large.
Mistral’s newly enhanced Le Chat, which translates to “the cat” in French, is receiving approximately six new features that elevate its functionality, transforming it into a more formidable work assistant and positioning it against competitors like ChatGPT and Claude from Anthropic PBC. One of the standout updates is the ability for Le Chat to conduct web searches and deliver answers complete with citations, mirroring the functionalities of ChatGPT and the generative AI-powered search engine Perplexity.
The new “Canvas” tool introduced for Le Chat draws parallels with ChatGPT’s Canvas feature, allowing users to edit, modify, and redesign content such as web pages and PowerPoint presentations using both text and voice commands. “You can use [the canvas feature] to create documents, presentations, code, mockups… the list goes on,” the company elaborated in a blog post unveiling these capabilities. “You’re able to modify its contents in place without regenerating responses, version your drafts, and preview your designs.”
Furthermore, Le Chat is now capable of ingesting and processing larger PDF documents, images, charts, and equations, enabling it to extract insights and summarize complex data more effectively. In a significant upgrade, Mistral confirmed that Le Chat’s image generation abilities have been enhanced through integration with Black Forest Labs Inc.’s Flux Pro model. This also allows Le Chat to facilitate automated workflows for intricate tasks such as invoice processing and expense reporting, effectively incorporating “agentic AI” to handle complex, multistep activities on behalf of users—an area where ChatGPT has yet to develop similar capabilities.
Most of the newly introduced features will be free to users while still in their beta phase, allowing broader access to these innovations. Among the new LLMs, Pixtral Large stands out as particularly noteworthy; it is a multimodal model engineered to process both textual and visual inputs. This marks the second installment in the Pixtral series, following the launch of the original Pixtral 12B in September.
With a monumental architecture of 124 billion parameters, Pixtral Large is a powerhouse, surpassing numerous leading multimodal models from competition such as Anthropic’s Claude 3.5 Sonnet, Google LLC’s Gemini 1.5 Pro, and OpenAI’s GPT-4o in several critical performance benchmarks. This substantial parameter count is essential, as it is a critical indicator of an LLM’s problem-solving capabilities—the higher the count, the superior the model’s performance.
“Pixtral Large is able to understand documents, charts, and natural images,” the company announced in a blog post on the launch of the model. “The model demonstrates frontier-level image understanding.” Sophia Yang, Mistral’s head of developer relations, noted that Pixtral Large excels particularly in areas such as multilingual optical character recognition, reasoning skills, and chart comprehension. To provide a practical demonstration, Yang shared a screenshot of Pixtral Large within Le Chat, showcasing its ability to analyze a restaurant bill using OCR technology to accurately allocate costs among diners.
Pixtral Large is equipped with an impressive context window of 128,000 tokens, allowing it to manage up to 30 high-resolution images at once or digest a 300-page book—capabilities that mirror those of OpenAI’s GPT-4o. This advanced model is readily available for download on Hugging Face, expanding access to its capabilities.
In addition to launching Pixtral Large, Mistral introduced the latest iteration of its flagship text-understanding model, Mistral Large. The newly minted version, known as Mistral Large 24.11, boasts enhancements in long context understanding, positioning it as even more effective for document analysis and related tasks.
Mistral stands among a wave of ambitious AI startups striving to challenge established leaders in the generative AI landscape, including OpenAI and Google. Established in April 2023 by a cohort of former employees from Google DeepMind and Meta Platforms Inc., Mistral has swiftly attracted considerable interest, leading to a valuation estimated at $6 billion following multiple high-profile funding rounds. To date, the startup has rolled out approximately a dozen AI models, catering to both commercial applications and research needs.
Notably, Mistral’s flagship models, including Mistral 7B, Mixtral 8x7B, and Mixtral 8x22B, have been released as open-source options, readily accessible via the Hugging Face platform. However, the Mistral Small, Medium, and Large models remain exclusive to users through the company’s application programming interface, accessible via licensing agreements.
Featured image: SiliconANGLE/Freepik AI; Mistral AI
Your vote of support is important to us and it helps us keep the content FREE. One click below supports our mission to provide free, deep, and relevant content. Join our community on YouTube, alongside more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger, and many more luminaries and experts.
“TheCUBE is an important partner to the industry. You guys really are a part of our events and we really appreciate you coming and I know people appreciate the content you create as well,” remarked CEO Andy Jassy.
THANK YOU
What innovative features does Mistral AI plan to introduce in the future to enhance user interaction with their models?
**Interview with Sophia Yang, Head of Developer Relations at Mistral AI**
**Interviewer:** Thank you for joining us today, Sophia. Mistral AI has made some impressive announcements recently, especially with the launch of your new models. Can you tell us more about the motivation behind developing the Pixtral Large model?
**Sophia Yang:** Absolutely! Our goal was to push the boundaries of what generative AI can do. With Pixtral Large, we wanted to create a multimodal model capable of processing textual and visual inputs simultaneously. This model not only surpasses existing offerings from industry giants but also enhances our understanding of documents, charts, and images. We believe that combining these capabilities will provide users with a more comprehensive understanding of complex data.
**Interviewer:** That’s fascinating! I understand that Pixtral Large boasts an impressive architecture with 124 billion parameters. How does this high parameter count translate to better performance?
**Sophia Yang:** Great question! A higher parameter count generally indicates a model’s enhanced problem-solving capabilities. With 124 billion parameters, Pixtral Large can analyze and interpret nuanced data more effectively, whether it’s understanding context in a text or providing insights from visual content. This gives us a significant edge over our competitors in critical performance benchmarks.
**Interviewer:** You also mentioned the enhanced features for your chatbot, Le Chat. How do these updates improve user experience?
**Sophia Yang:** Le Chat’s new features, such as the ability to conduct web searches and handle larger PDF documents, significantly elevate its functionality as a virtual assistant. The Canvas tool, allowing users to edit and design content with ease, offers a more interactive experience. It not only helps users to create and modify documents but also enables more complex tasks like invoice processing through integrated automated workflows.
**Interviewer:** It sounds like Mistral AI is making strides to compete with established names like ChatGPT. What are your expectations for the future of these models in the AI landscape?
**Sophia Yang:** We’re excited about the future! As we continue to enhance our models and introduce innovative features, our aim is to revolutionize the way people interact with AI. We plan to maintain our focus on accessibility as well, which is why most of our new features are available for free during the beta phase. We want as many users as possible to experience the power and versatility of Mistral AI.
**Interviewer:** Thank you, Sophia. It’s great to hear about Mistral AI’s vision and the advancements you are making in the field of generative AI.
**Sophia Yang:** Thank you for having me! We’re thrilled to share these innovations, and I look forward to seeing how users engage with our models.
