1
0
A French company called Mistral AI has entered the increasingly crowded AI market with a new proprietary large language model (LLM) called Mistral Large.
The company claims that Mistral Large can surpass several major competitors.
The Paris-based business said in a post on February 26 that Mistral Large performed well in multiple arithmetic and coding tests and outperformed many leading LLMs in a “multitask language understanding” test, with the exception of GPT-4.
Mistral Large’s performance, however, was not contrasted with that of Google’s Gemini Ultra or xAI’s Grok, which were both released in early February and November, respectively. Mistral AI has been contacted by Cointelegraph for comment.
The chief scientist and founder of Mistral AI, Guillaume Lample, thinks Mistral Large is “vastly superior” to Mistral AI’s earlier models.
Similar to how ChatGPT is based on GPT-3.5 and GPT, Mistral AI also published “Le Chat,” an AI chat interface built on top of its models.
The business, which said that Mistral Large is fluent in French, Spanish, German, and Italian, acquired US$487 million in funding in December from Nvidia, Salesforce, and Andreessen Horowitz.
Some viewers on X have expressed disappointment at Mistral Large, which is a closed, proprietary model similar to OpenAI’s current LLMs, despite the fact that Mistral AI’s original model was made available under an open-source license.
Read Also: AI at Your Service: How Advanced Package Tracking Solutions Are Changing the Game
Although Mistral Large has not been assessed by independent AI chatbot evaluation sites like Chatbot Arena, Mistral Medium, the company’s earlier model, is ranked sixth out of more than 60 LLMs.
A Bradley-Terry model processes the hundreds of pairwise scores in Chatbot Arena. It uses random sampling to produce a “Elo” rating, which indicates which model is most likely to prevail in a direct competition with another model.
The business and Microsoft have now announced a partnership that will allow Mistral Large to be accessed through Azure AI Studio and Azure Machine Learning.
Regarding its commercial LLM, Mistral AI said, “Microsoft’s trust in our model is a step forward in our journey.”
Eric Boyd, corporate vice president of Microsoft’s Azure AI Platform, stated on February 26 that Mistral Large will concentrate on AI research and development in addition to using Azure’s “supercomputing infrastructure” for training and scaling.
Mistral Large is slightly less expensive than GPT-4 Turbo at US$10 and US$30 per million tokens of input and output, respectively, costing US$8 per million tokens of input and US$24 per million tokens of output. Bloomberg estimates that Mistral’s value in December was close to US$2 billion.
Disclaimer: The information provided is for educational and informational purposes only. Always consult with relevant experts and professionals before making any decisions.
