French artificial intelligence startup Mistral announced the Mistral Large 2 model, its new generation flagship AI model. This model has 123 billion parameters and offers significant improvements in code generation, mathematics and reasoning capabilities over previous models. Additionally, it supports more than 80 programming languages and provides multi-language support.
Mistral introduces its Large 2 flagship AI model with 123 billion parameters
Mistral Large 2 is equipped with 128k context windows and supports multiple languages. The model reached an accuracy rate of 84.0 percent in the MMLU (Measured Misinformation and Lie Understanding) test. This new model includes significant improvements in code generation, logic execution and multi-language support.
Additionally, special efforts were made to minimize hallucination problems during training of the model. Mistral states that the Large 2 AI model has the ability to admit that it does not know, instead of lying about things it does not know. This increases the model’s ability to provide more reliable and accurate information.
Mistral Large 2 is available under an open license for non-commercial research purposes. Users can download the weights of the model and fine-tune it according to their needs. For commercial or corporate use, a separate license and usage agreement must be purchased from Mistral.
Although Mistral Large 2 has fewer parameters than the Llama 3.1 model, which has 405 billion parameters, it gives very close results in terms of performance. The model performs at the GPT-4 level and achieves very close results to the GPT-4o, Llama 3.1-405 and Anthropic’s Claude 3.5 Sonnet in various benchmark tests.
Mistral states that this new model will continue to increase cost effectiveness, speed and performance and offer new features for developing high-performance artificial intelligence applications for users.
What are you thinking? You can share your opinions in the comments section below.
Source link: https://shiftdelete.net/mistral-yeni-large-ai-2-modeli