French LLM wonder Mistral is getting ready to launch its next language model. You can already test it in chat.
The "Mistral Next" model has recently become available in direct chat mode in the chatbot arena. A Mistral developer confirmed on Discord that Mistral Next is the next prototype language model of the French LLM start-up. More details will follow soon.
Since the "Medium" model was recently released at GPT 3.5 level, there is reason to believe that "Next" could be the largest model of the startup so far, intended to catch up with GPT 4.
This is supported by information found in a Github repo of a French chat agent startup, where the new Mistral model is classified as "Large", while the Medium model has been downgraded from "Large" to "Medium".
The first testers on "X" are very positive about Mistral Next, suggesting that it could perform at GPT-4 level. You can test Mistral Next in Chatbot Arena's "Direct Chat".
Mistral is the darling of the open-source LLM scene
Mistral AI is a European AI startup that has taken an open approach to model development. It was founded by a team of leading researchers from Deepmind and Meta. Mistral AI's goal is to position itself at the forefront of generative AI research in Europe and to establish itself internationally.
Mistral AI has released the Mixtral 8x7B language model, an expert mixing model that has been very well received by the open-source community. This model uses a similar architecture to GPT-4 and is very efficient in terms of cost and latency.
In our content generation tests, Mixtral performs very well and can also execute GPT-4 prompts reliably, which is hard to say for Meta's Llama models, for example.
Mistral's 7B model is also a popular open-source language model for simpler applications that can outperform larger Llama models.
Mistral AI recently raised €385 million in a funding round led by Silicon Valley venture capital firm Andreessen Horowitz and early investor Lightspeed Ventures.