Helping to share the web since 1996


French AI Startup Mistral Unveils Mixtral 8x22B: A Frontier in Large Language Models

an iphone sitting on top of a table next to a leaf

French AI startup Mistral unveiled Mixtral 8x22B, a significant advancement in large language models (LLMs) as it eyes for dominance in the AI landscape. Expected to surpass Mistral’s previous Mixtral 8x7B LLM, early indications suggest it may even outshine competitors like OpenAI’s GPT-3.5 and Meta’s Llama 2, as reported by Gigazine.

The key features of the new Mixtral model include a substantial 65,000-token context window, allowing for comprehensive text processing and referencing, and a parameter size of up to 176 billion, reflecting the model’s vast internal variables for decision-making and predictions.

Developed by researchers from Google and Meta, Mistral adopts an open-source philosophy for its AI models. Mixtral 8x22B is freely available for use upon downloading a 281GB file, facilitated by pasting the magnet link from Mistral AI’s X post into a BitTorrent client of choice.

In a dynamic AI landscape marked by continuous innovation, Mistral’s latest LLM release coincides with other significant developments. OpenAI introduced GPT-4 Turbo with Vision, equipped to handle images alongside text, while Google launched Gemini Pro 1.5 LLM for developers, offering a free tier with limited daily requests. Meta’s forthcoming Llama 3 model further adds to the competitive atmosphere.

These advanced LLMs, including Mixtral 8x22B, are dubbed frontier models, designed to tackle diverse tasks with cutting-edge technology, reminiscent of the pioneering spirit of the Wild West. However, such advancements also raise regulatory concerns, as highlighted by OpenAI’s acknowledgment of the potential risks associated with frontier models in a July 2023 blog post.

Criticism of Mistral’s open-source strategy has emerged, notably regarding the startup’s inability to control the use and development of its AI models, as reported by The Guardian. Nonetheless, Mistral’s approach reflects a commitment to openness and collaboration in advancing AI technology.

«

»

Back to news headlines