Mistral Announces 1.05 Trillion Parameter Model Large 4
Mistral is positioning its largest open-weight release directly against Chinese open-weight models, and the company says it is deployable from European infrastructure through its own Mistral Cloud.
Reporting from 1 source: GIGAZINE.
Mistral announced Mistral Large 4, nicknamed Le Chonk, on October 6, 2026. The model has 1.05 trillion total parameters and 49 billion active parameters in a Mixture-of-Experts setup. It handles text and images, supports a 1-million-token context window and over 160 languages, and is available through the Mistral API, with open weights planned for the end of October, according to Mistral.
Mistral Large 4 carries 1.05 trillion total parameters but activates 49 billion during inference, a Mixture-of-Experts split that keeps the model's runtime cost tied to the smaller number. It also pairs that setup with a 1.6-billion-parameter image encoder and a 1-million-token context window, and Mistral says multilingual data covers more than 160 languages, including all official EU languages.
The company frames the model's benchmarks against Chinese open-weight competitors in particular. Mistral says its aggregated Coding Agent Index reached 49.8%, above DeepSeek V4 Pro 0813 and Qwen3.8 Max. In a code evaluation run with Surge AI, where professional scorers rated generated code without seeing model names, Mistral Large 4 averaged 3.74 out of 5.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.