Mistral vs Meta: Introducing Large 2

Just one day after Meta launched Llama 3.1, its biggest model to date, Microsoft-backed start-up, Mistral, launched its biggest model: Large 2, which builds on its predecessor and reportedly levels with GPT-4o and Llama 3.1 and beats Anthropic’s Claude 3.5 Sonnet in certain performance benchmark tests, namely coding and math.

Similar to Llama 3.1, Large 2 doesn’t have multimodal capabilities like GPT-4o does, but it is multilingual, understanding multiple languages, including English, French, German, Spanish, Italian, Portuguese, Arabic, Hindi, Russian, Chinese, Japanese, and Korean.

Like a lot of other AI models, it’s also not available for commercial use (unlike Llama 3.1) unless users pay for a license.

A common issue with its predecessor was its ability to generate code, so Mistral has rectified this by training the model so it understands and can generate code in over 80 programming languages, including Python, Java, C, C++, JavaScript, and Bash. As a result, it beat Llama 3.1 on code generation benchmark tests, although, GPT-4o pipped it to the number 1 spot. And it was a similar story for Math benchmark tests.

To reduce hallucinations Mistral has also fine-tuned the model to be more cautious and transparent when it doesn’t know the answer to a question, rather than making something up.

It offers more concise answers, is better at following instructions, can handle longer multi-turn conversations, and due to its larger context window, can process larger amounts of text, from a single prompt.