Mistral AI, Cohere raise funds to build advanced LLMs

A wave of startups is racing to build the next breakthrough in large language models (LLMs), targeting gaps left by today’s dominant systems. Companies like Mistral AI, Cohere, and Inflection AI are raising hundreds of millions and releasing models that challenge the performance of giants like OpenAI and Google. This surge comes as businesses rush to embed AI into products and services, creating demand for smarter, faster, and more specialized models.

The push began after Google researchers introduced the transformer architecture in 2017 with their paper “Attention Is All You Need.” That innovation enabled LLMs to learn from vast amounts of text and generate human-like responses. Since then, models have grown exponentially in size and capability, but they’ve also become expensive to train and run. Startups are now betting they can do better by focusing on efficiency, customization, or niche applications—like coding assistants, legal analysis, or healthcare chatbots. Mistral’s recent release of a 12.9-billion-parameter model, for example, reportedly matches older, larger models from top labs while using far less computing power.

Investors are pouring money into these efforts. Inflection AI closed a $1.3 billion funding round last year, Cohere raised $270 million in 2022, and Mistral recently secured $640 million. The startups argue their models offer faster response times, lower costs, or better performance on specific tasks. Anthropic and Meta have also entered the fray, releasing open-weight models to encourage adoption and experimentation. But competition is fierce, and many of these companies still trail in scale and ecosystem support compared to the biggest players.

What happens next may reshape how AI is deployed across industries. If these startups succeed, businesses could gain more control over their AI tools without relying solely on a handful of tech giants. That could mean faster innovation, lower costs, and more privacy-friendly options. But it also raises concerns about fragmentation, safety, and the concentration of AI power. As models improve and spread, the real test will be whether they can deliver real value beyond the hype—and who ends up using them.