Mistral AI CEO Announces Plans to Open-Source GPT-4 Level Model in 2024
-
Arthur Mensch, CEO of French startup Mistral AI, announced on French national radio that the company plans to open-source a GPT-4 level model in 2024, sparking industry attention.
Recently, Mistral AI launched an 8X7B model based on the MoE architecture, which rivals other popular models like GPT-3.5 and Llama2 70B. It outperforms Llama2 70B in most benchmarks with a 6x faster inference speed. The model is licensed under Apache 2.0, offering developers greater flexibility.
After the news spread, it sparked heated discussions on social media. One user expressed concerns about Mistral AI's future profitability, especially after the release of their so-called "Mistral Tiny" 7B model. This has also piqued curiosity about the upcoming "Mistral Medium" model.
Mistral AI categorizes its models into three tiers: Mistral Tiny, Mistral Small, and Mistral Medium. The currently released Mistral 7B Instruct v0.2 and Mixtral 8x7B belong to Mistral Tiny and Mistral Small, respectively. Interestingly, the Mistral Medium model has not been released yet, but the company stated it is under development and will rank among the top in standard benchmarks.
The company recently announced securing $415 million in funding at a $2 billion valuation, led by Andreessen Horowitz (a16z), with Lightspeed Venture Partners also participating in the follow-on investment. To address the commercial sustainability challenges faced by open-source large language model companies, Mistral AI recently launched 'La Plateforme,' providing API endpoints for its available models.
Additionally, Mistral AI is developing the multilingual Mistral Medium model, which excels in English, French, Italian, German, Spanish, and coding, achieving an 8.6 score on MT-Bench—surpassing even GPT-3.5.
Notably, rumors are circulating in the industry about OpenAI potentially releasing GPT-4.5 by year-end, making 2024 a year full of anticipation and competition in the open-source large language model space.