French startup Mistral AI released a chatbot and its flagship large language model (LLM) called Mistral Large, which gets pretty close to GPT-4 performance.
The meteoric rise of Mistral AI made headlines in December when it achieved a $2b valuation less than six months after being founded. It’s lightweight open-source Mixtral 8x7B LLM caused some excitement by beating bigger competitors like Meta’s Llama 2 34B.
There’s no detail on how many parameters Mistral Large has but the benchmark figures Mistral AI published show that its new model beats Claude 2, Gemini Pro, and Llama 2-70B. Its 32k context window is smaller than some of those models but it’s still equivalent to around 20,000 words.
OpenAI’s GPT-4 continues to be the golden standard for comparing LLM performance and Mistral Large comes pretty close to matching it in a number of benchmarks.
The pricing model for API access to Mistral Large is another shot across OpenAI’s bow. Querying Mistral Large costs $8 per million input tokens and $24 per million output tokens. That’s 20% cheaper than what OpenAI charges to query GPT-4.
While Mistral AI is a big supporter of open-source models, its new models have been released with API access only, unlike Mixtral 8x7B.
Le Chat
Mistral AI also announced a beta release of its chatbot called Le Chat. The conversational chatbot allows users to experience responses from Mistral Large but also offers access to Mistral AI’s new smaller models, Mistral Small and Mistral Next.
The latter two LLMs offer more concise and cheaper interactions. Mistral Small outperforms Mixtral 8x7B and has lower latency.
Mistral AI says it has tried to make its models “as useful and as little opinionated as possible.” The company says that with its “tunable system-level moderation mechanism, le Chat warns you in a non-invasive way when you’re pushing the conversation in directions where the assistant may produce sensitive or controversial content.”
So, expect a little less wokeness than Gemini may deliver. You can try Le Chat out here.
Microsoft endorsement
Even though Microsoft is the biggest financial backer of OpenAI, it has made several other models available on its platforms. Coinciding with the release of Mistral Large, Microsoft announced it was entering a multi-year partnership with Mistral AI and that it would make its Azure platform available for Mistral AI to develop and deploy its models.
Mistral AI CEO Arthur Mensch said, “We are thrilled to embark on this partnership with Microsoft. With Azure’s cutting-edge AI infrastructure, we are reaching a new milestone in our expansion propelling our innovative research and practical applications to new customers everywhere.”
With the release of new Gemini models and Mistral Large hot on the heels of GPT-4, you’ve got to wonder if the release of GPT-5 is imminent. If Mistral AI adds multimodal functionality to Mistral Large then the pricing makes it a big threat to OpenAI.
