Mistral Releases Large 4, a 1 Trillion Parameter AI Model, as Open-Weight Competition With US Labs Intensifies

French AI company Mistral released Mistral Large 4 in October 2026, a 1 trillion parameter Mixture-of-Experts model that the company says competes directly with the leading closed AI systems from OpenAI and Anthropic.

The release marks a significant step in the ongoing competition between open-weight and proprietary AI models. Mistral has positioned itself as a European alternative to US-based AI labs, emphasizing transparency and accessibility as core values.

Mixture-of-Experts architecture allows the model to activate only a subset of its parameters for any given task, making it more computationally efficient than a dense model of comparable size. Mistral said this design allows Large 4 to deliver strong performance without requiring the same level of computing resources as some competing systems.

Early benchmark results shared by Mistral showed the model performing competitively on coding, reasoning, and multilingual tasks. Independent researchers said they would need time to verify the claims through their own testing.

The release comes as the US startup Reflection AI also introduced Beam, a 501 billion parameter open-weight model designed to compete with Chinese AI infrastructure. Together, the two releases signal that the open-weight segment of the AI market is becoming more competitive with proprietary systems.

Mistral Large 4 is available through the company's API and can be downloaded and run locally by organizations with sufficient computing resources. The company said it plans to release additional documentation and fine-tuning guides in the weeks following the launch.

European regulators have been watching Mistral's growth closely. The company has received significant investment from European venture funds and has been cited as an example of the continent's ability to compete in the global AI race.