Mistral lança o Pixtral Large e atualiza o Le Chat com geração de imagens.

Sure! Here’s a rewritten version of the content while maintaining the HTML tags:

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More

<p>Mistral, a French startup that gained significant attention last year with record-breaking seed funding in Europe, has rolled out a series of updates today, including a new large foundational model called Pixtral Large.</p>

<p>The company is enhancing its free web-driven chatbot, Le Chat, by adding features like image generation, web search, and an interactive "canvas," which aligns its offerings more closely with OpenAI's ChatGPT, making it a strong competitor in the space.</p>

<p>As Mistral AI's CEO and co-founder, Arthur Mensch, stated on his account on the social platform X, “At Mistral, we recognize that to deliver the best AI experience, one must collaboratively design models and product interfaces. Pixtral was trained with impactful front-end applications in mind and represents an excellent example of this approach.”</p>

<p>Users interested in exploring the new Le Chat features will need to enable them as beta features on the web interface. Please be aware that access to Le Chat requires a free account from Mistral, Google, or Microsoft.</p>

<h2>Pixtral Large — Open Source Multimodal AI</h2>

<p>Pixtral Large, Mistral’s new model with 124 billion parameters, builds upon its predecessor, Mistral Large 2, launched during the summer of 2024, as well as the first multimodal model, Pixtral 12-B, introduced in September.</p>

<p>This model incorporates a 123-billion-parameter decoder alongside a 1-billion-parameter vision encoder, making it proficient in both text and visual data processing.</p>

<p>Parameters indicate the number of settings that control a model's inputs and outputs; thus, a higher number of parameters typically suggests a more capable, knowledgeable, and efficient model.</p>

<p>According to a post by Mistral's Head of Developer Relations, Sophia Yang, on her X account, Pixtral Large excels at “multilingual OCR [optical character recognition], reasoning, chart understanding, and more.” Yang shared a screenshot showcasing Pixtral Large analyzing a receipt uploaded by a user via OCR, exemplifying its capabilities in ingesting and documenting expenses, including splitting a bill with a tip.</p>

<p>With a context window of 128,000 tokens, Pixtral Large can process up to 30 high-resolution images per input or approximately a 300-page book, which is comparable to leading models from OpenAI's GPT series.</p>

<p>The model showcases state-of-the-art performance across various benchmarks such as MathVista, DocVQA, and VQAv2, making it well-suited for tasks like chart interpretation, document analysis, and image comprehension.</p>

<p>While the model and its weights can be downloaded for free from Hugging Face, they are available under a custom Mistral AI Research License, which restricts use to non-commercial, research purposes. Organizations seeking commercial utilization will need to access it through Mistral’s API on its Le Platforme managed web service or obtain a separate license directly from the company, indicating that it is not entirely open source.</p>

<p>Nonetheless, by providing Pixtral Large, Mistral AI enables researchers and developers to leverage advanced multimodal AI while promoting responsible and ethical usage.</p>

<h2>Le Chat Competes with ChatGPT's Comparable Features</h2>

<p>At the core of Mistral’s AI suite is Le Chat, a free platform now upgraded with new functionalities courtesy of Pixtral Large.</p>

<p>Tailored for various applications such as research, ideation, and automation, Le Chat seamlessly integrates text, vision, and interactive features into a cohesive productivity toolkit.</p>

<h3>New Features of Le Chat:</h3>

<ol>
<li><strong>Web Search with Citations:</strong> Users can enhance the AI’s knowledge base with real-time web searches, complete with source citations for improved transparency.</li>

<li><strong>Canvas for Ideation:</strong> This innovative interface permits users to create, edit, and collaborate on documents, presentations, and designs in a new interactive space appearing to the left of the chatbot interface. As Yang noted on X, Le Chat Canvas is “ideal for creative ideation. You can utilize Canvas to forge documents, presentations, code, mockups… the possibilities are endless.” This feature arrives just six weeks after OpenAI unveiled its own Canvas sidebar for ChatGPT, seen by many as a direct response to Anthropic’s earlier Artifacts feature for its Claude chatbot.</li>

<li><strong>Advanced Document and Image Analysis:</strong> With Pixtral Large, Le Chat can efficiently analyze and summarize intricate PDFs, distilling insights from graphs, tables, equations, and more.</li>

<li><strong>Image Generation:</strong> Through a partnership with Black Forest Labs, Le Chat now boasts image generation capabilities powered by the Flux Pro model, enabling users to create high-quality visuals directly within the chat interface. This aligns as a competitive response to OpenAI’s DALL-E 3 integration in ChatGPT and marks the second major incorporation of Black Forest Labs' new models into a leading AI foundation model provider's offerings, following its collaboration with Elon Musk’s xAI for image generation within that company’s Grok-2 chatbot on the X platform, which Musk also owns.</li>

<li><strong>Task Agents for Automation:</strong> Customizable agents can automate repetitive tasks such as summarizing meeting notes, processing invoices, or scanning receipts, thereby conserving users' time and effort.</li>
</ol>

<p>These additions position Le Chat as a multifunctional AI assistant, adept at managing tasks that traditionally necessitate several tools.</p>

<p>Mistral AI underscores Le Chat’s extensive feature set alongside its accessibility relative to platforms like ChatGPT, Perplexity, and Claude. While competitors may impose premium subscriptions for similar capabilities, Le Chat offers a fully integrated, multimodal experience free of charge during its beta phase.</p>

<h2>Mistral is Stepping Up Its Game</h2>

<p>With the launch of Pixtral Large and the enhanced Le Chat, Mistral is showcasing its research and development strength.</p>

<p>Despite some experts in the tech industry suggesting that the costs associated with intelligence are decreasing, complicating efforts for model providers to generate revenue, Mistral remains committed to advancing its offerings to compete with other industry leaders, even with a parameter count of 124 billion as opposed to Meta's latest Llama 3.1 release, which boasts 405 billion parameters.</p>

<p>However, Mistral has yet to incorporate advanced voice and audio capabilities found in competitors’ offerings like OpenAI’s ChatGPT Advanced Voice Mode or Google’s Gemini Live.</p>

<p>A recent survey by Kong indicated that despite Mistral's technical abilities and a mix of open-source and proprietary offerings, the utilization of Mistral's models and APIs by large enterprises lags significantly behind that of US-based companies such as OpenAI, Anthropic, and Microsoft.</p>

<p>Nevertheless, with the upcoming presidential election and the influence of xAI founder Elon Musk over President Trump, it's probable that the EU and its members will look to Mistral as a pathway for AI development outside the purview of the United States and its new, contentious leadership.</p>

<p>In other words, AI is swiftly becoming intertwined with nationalism and geopolitics, and Mistral finds itself in a potentially advantageous position as one of Europe's foremost AI model providers.</p>

<div>
    <h3>VB Daily</h3>
    <p>Stay informed! Receive the latest news in your inbox each day.</p>
    <p>By subscribing, you agree to VentureBeat's Terms of Service.</p>
    <p>Thank you for subscribing. Explore more VB newsletters <a href="#">here</a>.</p>
    <p>An error occurred.</p>
</div>

If you need any further modifications, just let me know!

---