Mistral AI, a Paris-based large language model startup, is reportedly in talks with investors to raise capital at a $5 billion valuation.

The Information reported the fundraising push on Tuesday evening, citing a person with direct knowledge of the matter. It’s believed Mistral could raise several hundred million dollars through the round. It’s unclear which investors might participate.

The report of the new funding round follows a year in which Mistral closed four investments collectively worth over $500 million. The company raised its first funding round, a $113 million seed raise, last June about a month after launching. A $415 million investment in December boosted Mistral’s valuation to $2 billion.

The LLM developer went on to raise $16.3 million from Microsoft Corp. in February and, most recently, an undisclosed sum from Databricks Inc. last month. Both investments were made in connection with product partnerships centered on Mistral’s LLMs. As part of those partnerships, the AI developer will make several of its models available through Databricks’ analytics platform and Azure.

Mistral’s newest open-source LLM, Mixtral 8x22B, debuted last week. In an internal evaluation, the company compared the model with Meta Platforms Inc.’s Llama 2 70B using a half dozen math and coding benchmark tests. Mixtral 8x22B scored significantly higher across all six assessments.

The LLM is based on a so-called mixture-of-experts architecture. It’s not a single, monolithic AI model but rather a collection of neural networks that are each optimized for a different set of tasks. When Mixtral 8x22B receives a prompt, it routes the request to the neural network best equipped to generate a response.

One major benefit of the mixture-of-experts approach is that it can reduce AI models’ hardware usage. When an LLM based on the architecture receives a prompt, it typically uses only one of the neural networks it includes to generate an answer. The others remain inactive and consequently don’t use any infrastructure resources.

Mixtral 8x22B’s competitive performance may be one reason Mistral believes it can secure a $5 billion valuation, which is more than double the $2 billion it was worth in December. There may also be other considerations behind the potential valuation boost. According to The Information, Mistral recently began generating revenue.

The company debuted its first commercial offerings on the same February day it announced its partnership with Microsoft. The product portfolio is headlined by Mistral Large, an LLM touted as the most advanced of its kind on the market behind GPT-4. The company also offers a more cost-efficient language model, Mistral Small, as well as a ChatGPT-like enterprise chatbot.

OpenAI is reportedly gearing up to launch a new flagship LLM called GPT-5. To keep pace with the company and other well-funded startup rivals, Mistral will likely have to make significant investments in new LLM development. The additional capital it’s reportedly seeking to raise could advance the effort. 

Photo: Mistral