When it rains, it pours for frontier AI models. Mistral launched its new flagship model, Large 2, on Wednesday, which it claims is on par with the latest cutting-edge models from OpenAI and Meta in terms of code generation, math, and inference.
The release of the Mistral Large 2 comes just a day after Meta released its latest and greatest open source model, the Llama 3.1 405b. Mistral says the Large 2 raises the bar for performance and cost for open models, and backs it up with some benchmarks.
Large 2 is shown to be faster than Llama 3.1 405B in both code generation and mathematical performance, and achieves this with less than a third of the parameters – 123 billion to be exact.
Mistral said in a press release that one of its main focus areas during training was minimizing the model’s hallucination problem. The company says Large 2 was trained to be more discriminating in its responses, and to admit when it doesn’t know something instead of just making something up that sounds plausible.
The Paris-based AI startup recently raised $640 million in a Series B funding round led by General Catalyst, at a valuation of $6 billion. Mistral is one of the new entrants to the AI space, but is quickly releasing state-of-the-art or near-state-of-the-art AI models.
However, it is important to note that Mistral’s model, like most others, is not open source in the traditional sense. Commercial use of the model requires a paid license. While it is more open than GPT-4o, there are very few people in the world with the expertise and infrastructure to implement such a large model. (Of course, this also applies to Llama’s 405 billion parameters.)
One thing that was missing from Mistral Large 2, and also from Meta’s Llama 3.1 release yesterday, was multimodal capabilities. OpenAI is way ahead of the game when it comes to multimodal AI systems, being able to process images and text simultaneously, a capability that some startups are increasingly looking to build.
The model has a 128,000 token window, which means Large 2 can process a lot of data in a single prompt (128,000 tokens is equivalent to about a 300-page book). Mistral’s new model also includes improved multilingual support. Large 2 understands English, French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese, Korean, and 80 other coding languages. Mistral claims Large 2 produces more concise responses than leading AI models, which tend to babble.
Mistral Large 2 is available on Google Vertex AI, Amazon Bedrock, Azure AI Studio, and IBM watsonx.ai. You can also access the new model on Mistral’s le Plateforme under the name “mistral-large-2407,” or try it out for free on the startup’s ChatGPT competitor, le Chat.
