Highlights:

  • The LLM developer proceeded to raise USD 16.3 million from Microsoft Corp. in February and, most recently, secured an undisclosed amount from Databricks Inc. last month.
  • When an LLM utilizing this architecture receives a prompt, it usually uses only one of the neural networks it incorporates to produce an answer.

A Paris-based LLM developer, Mistral AI to raise USD s billion in the latest funding. The startup is in discussions with potential investors.

Reports revealed the fundraising endeavor recently. It’s speculated that Mistral AI funding could raise several hundred million dollars through the round. The participating investors remain uncertain at this time.

The report of the new funding round comes after a year during which Mistral secured four investments totaling over USD 500 million in value. The company initiated its first funding round, securing a USD 113 million seed raise last June, approximately a month after its launch. Following a USD 415 million investment in December, Mistral’s valuation soared to USD 2 billion.

The LLM developer proceeded to raise USD 16.3 million from Microsoft Corp. in February and, most recently, secured an undisclosed amount from Databricks Inc. last month.
Both investments were made in association with product partnerships focused on Mistral’s LLMs. As part of those partnerships, the AI developer will offer several models through Azure and Databricks’ analytics platform.

Mistral’s latest open-source LLM, Mixtral 8x22B, was introduced last week. In an internal assessment, the company evaluated the model against Meta Platforms Inc.’s Llama 2 70B using half a dozen calculations and coding benchmark tests. Mixtral 8x22B outperformed significantly across all six evaluations.

The LLM is founded on a mixture of expert architecture. It’s not a singular, monolithic AI model; instead, it comprises a collection of neural networks optimized for various tasks. When prompted, Mixtral 8x22B directs the request to the neural network most suitable for generating a response.

One significant advantage of the mixture-of-experts approach is its potential to decrease hardware usage in AI models. When an LLM utilizing this architecture receives a prompt, it usually uses only one of the neural networks it incorporates to produce an answer. The remaining neural networks remain inactive and thus do not consume any infrastructure resources.

Mistral’s confidence in securing a USD 5 billion valuation, more than doubling its worth from USD 2 billion in December, may be partly attributed to the competitive performance of Mixtral 8x22B. There may be additional factors contributing to the potential valuation increase. According to the reports, Mistral has recently started generating revenue, which could positively impact its perceived value.

The company launched its initial commercial offerings on the same day in February when it unveiled its partnership with Microsoft.

The flagship product in Mistral’s portfolio is Mistral Large, an LLM hailed as one of the most advanced models on the market, second only to GPT-4. In addition to Mistral Large, the company provides a more cost-efficient language model, Mistral Small. Furthermore, it offers an enterprise chatbot akin to ChatGPT.

OpenAI is purportedly preparing to introduce a new flagship LLM, GPT-5. Mistral will likely need to make substantial investments in developing new LLMs to remain competitive with the company and other well-funded startup rivals.
The additional capital that Mistral reportedly seeks to raise could facilitate its efforts to advance.