- Mistral unveiled its latest flagship AI model, Large 2, positioning it as a formidable competitor to cutting-edge models from OpenAI and Meta on July 24th.
- The release of Mistral’s Large 2 marks a significant milestone in the rapidly evolving AI landscape, highlighting several critical aspects of current AI development and competition.
OUR TAKE
Mistral’s Large 2 makes a strong debut, challenging Meta and OpenAI’s giants. However, size isn’t everything. With 123 billion parameters compared to Llama’s 405 billion, Large 2’s precision over brute force is intriguing. Yet, its closed-source nature and expensive licensing could be off-putting for some. Additionally, the lack of multimodal capabilities means OpenAI still leads in image-text integration. Mistral’s rapid rise, combined with its 128K token window, demonstrates their strategic approach, advancing one book-sized prompt at a time.
–Miurio huang, BTW reporter
What happened
Mistral AI unveiled its latest flagship AI model, Large 2, positioning it as a formidable competitor to cutting-edge models from OpenAI and Meta on July 24th. This release came just a day after Meta introduced its own advanced model, Llama 3.1 405b. Mistral asserts that Large 2 excels in areas such as code generation, mathematics, and reasoning, matching or surpassing its rivals in performance.
Large 2 boasts 123 billion parameters, significantly fewer than Llama 3.1 405B, yet it reportedly outperforms Meta’s model in specific benchmarks. A primary focus during Large 2’s training was to reduce hallucination issues, with the model designed to acknowledge when it lacks knowledge rather than fabricating plausible-sounding information.
The Paris-based AI startup, Mistral, recently raised $640 million in a Series B funding round, led by General Catalyst, achieving a valuation of $6 billion. Despite being a relatively new player in the AI industry, Mistral has quickly advanced to the forefront, consistently releasing high-performance AI models.
However, it is crucial to note that Mistral’s models, like many others, are not open source in the traditional sense. Commercial applications of Large 2 require a paid license, and implementing such a large model demands substantial expertise and infrastructure. Additionally, Large 2, similar to Meta’s Llama 3.1, lacks multimodal capabilities—a feature where OpenAI currently leads, enabling AI to process images and text simultaneously.
Large 2 features a 128,000-token window, allowing it to handle substantial data in a single prompt, equivalent to about a 300-page book. The model also offers enhanced multilingual support, understanding languages such as English, French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese, and Korean, as well as 80 coding languages. Notably, Mistral claims Large 2 provides more concise responses than other AI models, which often tend to be verbose.
The new model is accessible on major platforms like Google Vertex AI, Amazon Bedrock, Azure AI Studio, and IBM Watsonx. Additionally, users can explore Large 2 on Mistral’s platform under the name “mistral-large-2407” and test it for free on the startup’s ChatGPT competitor, IE Chat.
Also read: UK leads Europe in GenAI startups, Accel reveals
Also read: The difference between Conversational AI and GenAI
Why it’s important
The release of Mistral’s Large 2 marks a significant milestone in the rapidly evolving AI landscape, highlighting several critical aspects of current AI development and competition. Firstly, the model’s performance benchmarks and parameter efficiency underscore the ongoing race among AI companies to create more powerful and cost-effective solutions. By outperforming Meta’s Llama 3.1 405B with fewer parameters, Mistral demonstrates that efficiency and performance can coexist, setting a new standard for AI models.
Reducing hallucination issues in AI models is another crucial advancement. AI systems that can accurately acknowledge their limitations enhance reliability and trustworthiness, particularly in professional and critical applications. This focus on creating a discerning model positions Mistral as a leader in developing more dependable AI solutions.
The substantial Series B funding and the $6 billion valuation reflect the high level of investor confidence in Mistral’s capabilities and market potential. This financial backing will likely fuel further innovation and expansion, enabling Mistral to continue challenging established AI giants like OpenAI and Meta.
However, the proprietary nature of Large 2, requiring paid licenses for commercial use, highlights a broader industry trend towards monetising advanced AI technologies. This approach can limit accessibility for smaller enterprises or individuals lacking the necessary resources, potentially slowing widespread adoption and innovation.
The lack of multimodal capabilities in Large 2 points to an area where OpenAI maintains a competitive edge. Multimodal AI systems are increasingly important for applications requiring the simultaneous processing of diverse data types, such as images and text. As startups strive to incorporate these features, the competition will likely intensify, driving further advancements in AI technology.
Large 2’s extensive multilingual support and ability to process a large volume of data in a single prompt make it a versatile tool for global applications. This capability is particularly valuable in diverse linguistic environments and complex computational tasks, broadening the model’s applicability across various sectors.
Mistral’s Large 2 represents a significant leap forward in AI development, showcasing the company’s ability to compete with industry leaders. The model’s advancements in performance, reliability, and multilingual support position Mistral as a formidable player in the AI landscape, driving innovation and setting new benchmarks for future AI models.
