Mistral Shakes Up the AI Landscape with the Release of Large 2
In the fast-paced world of artificial intelligence, the competition is fierce. Mistral, a Paris-based AI startup, has just unveiled its latest flagship model, Large 2. The company claims that Large 2 is on par with the cutting-edge models released by OpenAI and Meta in terms of code generation, mathematics, and reasoning. This release comes hot on the heels of Meta’s Llama 3.1 405b model, creating quite a buzz in the AI community.
Mistral’s Large 2 boasts impressive performance and cost efficiency when compared to Meta’s Llama 3.1 405B. Surprisingly, Large 2 outperforms Llama on code generation and math performance while utilizing less than a third of the parameters, with a precise count of 123 billion.
One of Mistral’s key focuses during the training of Large 2 was to address its hallucination issues. The company claims that the model has been trained to be more discerning in its responses, avoiding the creation of plausible but inaccurate information. Large 2 is designed to acknowledge when it lacks knowledge on a particular topic instead of fabricating responses.
Despite being one of the newer players in the AI space, Mistral has quickly gained attention and credibility. The startup recently secured $640 million in a Series B funding round led by General Catalyst, valuing the company at $6 billion. Mistral’s ability to deliver AI models on or near the cutting edge has contributed to its rapid growth and success.
However, it’s important to note that Mistral’s models are not open source in the traditional sense. A paid license is required for any commercial use of the model. Additionally, implementing such large-scale models requires significant expertise and infrastructure, making them inaccessible to many organizations.
One notable feature missing from both Mistral’s Large 2 and Meta’s Llama 3.1 release is multimodal capabilities. OpenAI currently leads the pack in this area with its ability to process both images and text simultaneously. Nevertheless, startups are increasingly looking to develop multimodal AI systems, recognizing the value and potential of combining these two modalities.
Mistral’s Large 2 comes with a 128,000 token window, allowing it to process a substantial amount of data in a single prompt. To put it into perspective, this is equivalent to roughly a 300-page book. Additionally, the model offers improved multilingual support, with proficiency in twelve languages and 80 coding languages. Mistral claims that Large 2 produces more concise responses compared to leading AI models, which have a tendency to ramble on.
In terms of accessibility, Mistral has ensured that Large 2 can be used on popular platforms such as Google Vertex AI, Amazon Bedrock, Azure AI Studio, and IBM watsonx.ai. Additionally, the model is available on Mistral’s own platform, le Plateforme, under the name “mistral-large-2407.” Users can also test out Large 2 for free on Mistral’s ChatGPT competitor, le Chat.
Mistral’s release of Large 2 has undoubtedly made a significant impact in the AI sphere. With its impressive performance, cost efficiency, and multilingual support, Large 2 presents a compelling option for organizations seeking cutting-edge AI capabilities. While challenges remain in terms of accessibility and multimodal capabilities, Mistral’s entrance into the market has solidified its position as a serious contender in the AI landscape.
