Mistral, the artificial intelligence company, has recently unveiled two new models without divulging their internal structure for public use through the cloud. These two models are:
“Mistral Large,” the company’s most advanced model, supports English, French, Spanish, German, and Italian languages, with a data capacity of 32K tokens. It can be customized to execute commands precisely and allows developers to personalize content filtering. The general knowledge test score of this model is second only to GPT-4.
On the other hand, “Mistral Small” surpasses the efficiency of Mixtral 8x7B that the company previously made available for public use. It emphasizes a similar high speed performance.
Both models have been adjusted to support output generation in JSON format and can call external APIs. Mistral provides customers with three options to use the company’s closed models: 1) Direct usage from Mistral, 2) Utilization through Microsoft’s Azure via the Azure Marketplace, and 3) Contact for purchasing and installing the models within the organization. The pricing for Mistral Large is $8 per 1 million input tokens and $24 per 1 million output tokens.
Currently, Mistral offers a total of 5 models for service, including two open models – Mistral 7B and Mixtral 8x7B, and three closed models – Mistral Small, Mistral Medium, and Mistral Large.
TLDR: Mistral has introduced two new AI models, Mistral Large and Mistral Small, both of which support multiple languages and offer efficient performance with options for direct usage or integration through Azure.
Leave a Comment