anticipated 405-billion-parameter version of Llama 3. Mistral states Large 2 boosts the bar for performance and cost for open models. The model supports a couple of benchmarks.
Mistral AI is a French company specializing in artificial intelligence (AI) products. The company was founded in 2023 by former Meta Platforms and Google DeepMind employees. The company focuses on producing large language models with open sources. They also accentuate the importance of free and open-source software and locate themselves as an alternative to proprietary models. Mistral is in fourth place in the global AI race and aims to democratize AI by concentrating on open-source innovation.
The new release of Mistral, Large2, overtakes Llama 3.1 405B on code generation and math performance. If Mistral’s benchmarks are veritable, Large2 fights with OpenAI’s GPT-4o, Anthropic’s Claude 3.5 Sonnet, or other AI models. In the popular Massive Multitask Language Understanding (MMLU) benchmark, Large 2 achieves a score of 84 percent. In contrast, Meta’s new release achieved a score of 88,6 percent, while OpenAi GPT-4o managed a score of 88,7 percent.
The main factor is that Large2 handles achieve this level of performance using a few resources competing models. Large2 is smaller than Meta’s oldest model and almost one-fourth the magnitude of GPT-4. These implications can make Large 2 one of the most attractive models of commercial applications.
Mistral said that reducing the model’s hallucination issues was a key focus in training. Large2 was skilled in being more intelligent in responses. It would recognize a mistake instead of making up something that is plausible. The artificial intelligence startup recently increased $640 million in a Series B funding round led by General Catalyst.
Mistral Large2 and also Meta’s Llama 3 have some losses. The multimodal capabilities are missing. OpenAI is far ahead in multimodal artificial intelligence systems that can process images and text at the same time. Mistral’s new release provides multilingual support. Large 2 understands English, French, German, Arabic, Chinese, and other languages. It also knows 80 coding languages.
Large2 is accesible on Google Vertex AI, Amazon Bedrock, Azure AI Studio, and IBM Watsonx.ai. The new model is on Mistral’s Platform under ”mistral-large-2407”.
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
